Nested models

The pieces the models above are built from.

Alphabetical. Each says where it is used.

Art

Artwork a style renders in its own medium: ascii-art prints it in characters, particles fills the screen with it in points. Generated locally from a prompt (or taken from a file), optionally animated into a short loop.

Field	Type	Default	Description
prompt	string (optional)	null	Generate the image locally (Z-Image), e.g. 'a basketball player dribbling, black background'. Plain backgrounds and strong silhouettes read best in characters and points
file	string (optional)	null	Use this image (project file or URL) instead of a prompt
motion	string (optional)	null	Animate it with the video model (needs ComfyUI), e.g. 'dribbles the ball between his legs'. Omit for a still
seconds	number	3.0	Length of the animated loop Constraints: 1 to 8.
size	list of 2 integers	[1024, 768]	Generated image size (the loop keeps its aspect) Constraints: exactly 2 items.
seed	integer	7
frames	integer	36	Frames shipped to the page (sampled evenly from the loop) Constraints: 1 to 96.
width	integer	220	Sampling width in pixels of the shipped frames: characters and points are drawn from these Constraints: 40 to 480.
label	string (optional)	null	Short name shown with the art (defaults to the prompt)
video_model	enum (optional)	null	Video model: ltx (LTX-2.3, the default) or wan (Wan 2.2 A14B: renders at ~1280x720 and 16 fps, retimed to 24; clean, physical motion and first/last-frame control, but no finer detail than LTX after restore, and slower). Unset = the site's video_model, then studio.toml's, then ltx Allowed values: "ltx", "wan".

Used in: Theme.art.

AtBox

A box on the stage, in fractions of its width and height (0 is the left or top edge, 1 the right or bottom).

Field	Type	Default	Description
x	number	required	Where the box's left edge is, 0-1 across the stage Constraints: 0 to 1.
y	number	required	Where the box's top edge is, 0-1 down the stage Constraints: 0 to 1.
w	number	0.34	How wide the box is, 0-1 of the stage's width Constraints: at most 1.
h	number (optional)	null	How tall the box is, 0-1 of the stage's height (default: as tall as its words) Constraints: at most 1.
align	enum	"left"	How the words line up inside the box Allowed values: "left", "center", "right".

Used in: Step.at.

AtGrid

A slot on the theme's 12-column grid: the content width between the page's gutters, cut into 12 columns.

Field	Type	Default	Description
grid	string	required	The columns, written as CSS writes them: "9 / span 4" (start at column 9, four columns wide), "1 / 7" (columns 1 to 6) or "5" (column 5 alone)
row	enum	"middle"	Where the slot sits down the stage Allowed values: "top", "middle", "bottom".
align	enum	"left"	How the words line up inside the slot Allowed values: "left", "center", "right".

Used in: Step.at.

Badge

A small pill above a headline: 'Brand Voices is here', optionally a link, with a dot or a tag in front.

Field	Type	Default	Description
text	string	required	The words in the pill
href	string (optional)	null	Make the pill a link (an arrow follows the words)
tag	string (optional)	null	A short filled label before the words, e.g. 'New'
icon	enum (optional)	null	A mark before the words: dot (a small accent dot) or spark Allowed values: "dot", "spark".

Used in: HeroBlock.badge, Step.badge.

Beat

One shot of a story film: a still generated from prompt (or taken from image), brought to life by motion.

Field	Type	Default	Description
id	string	required
prompt	string	""	The opening frame, described like a photograph: subject, place, light, lens (not used by a frozen moment)
motion	string	""	What moves and how the camera moves, e.g. 'the door swings open, the camera pushes in' (not used by a frozen moment)
seconds	number	4.0	Length of the shot (1 to 10 generated; a frozen moment may run longer) Constraints: 1 to 60.
seed	integer	7
image	string or ImageInput (optional)	null	Start from this image instead of generating it from prompt: a file, or an image input (generate:, edit: ...)
end_image	string or ImageInput (optional)	null	The clip's last frame (same forms as image): the motion is steered to land on it, e.g. a closed door ending open
strength	number	1.0	How tightly the clip holds the start image. Lower it (0.7-0.85) when an action shot freezes on its first frame Constraints: 0.3 to 1.0.
motion_seed	integer (optional)	null	Seed for the motion only, to re-roll a clip and keep its still
trim	number	0.0	Seconds cut from the start of the clip (an action shot that holds its first frame before moving). Re-uses the render: nothing is re-animated Constraints: at least 0.
fade	number (optional)	null	The join into this beat from the one before, in seconds of crossfade: 0 is a hard cut (a change of angle, a jump to a new moment), more is a longer dissolve. Default: the film's beat_fade. Not used on the first beat, or on a camera beat that continues the one before (that join is always frame-exact) Constraints: 0 to 2.
video_model	enum (optional)	null	Video model: ltx (LTX-2.3, the default) or wan (Wan 2.2 A14B: renders at ~1280x720 and 16 fps, retimed to 24; clean, physical motion and first/last-frame control, but no finer detail than LTX after restore, and slower). Unset = the site's video_model, then studio.toml's, then ltx. strength applies to ltx only Allowed values: "ltx", "wan".
locked	boolean	false	Keep this beat's clip (and its storyboard still) as rendered, even when its prompt or motion changes; the rest of the film still builds around it
freeze	Freeze (optional)	null	A frozen moment instead of a generated clip: one frame, upscaled, with a 2.5D camera moving through it, rendered offline in seconds (see Freeze). prompt, motion, image and end_image are not used
camera	CameraMove or CameraPreset or enum (optional)	null	A camera move on this moving beat: a named move (camera: dolly-in, or {move: orbit-left, amount: 0.8}, see CameraPreset) or keys (see CameraMove): camera: {keys: [{t: 0, box: [0,0,1,1]}, {t: 0.5, box: [...]}, ...]}. The crowd keeps moving, the camera does what the move or the keys say. Needs an image (or a prompt to make one); not for a frozen moment Allowed values: "static", "dolly-in", "dolly-out", "crash-zoom", "dolly-zoom", "orbit-left", "orbit-right", "crane-up", "crane-down", "fpv", "whip-pan", "push-through".
restore	integer (optional)	null	Run SeedVR2 over this beat's clip at this height (e.g. 1088) to clean grain and redraw fine detail, then scale it back to beat_res. Like the section's beat_restore but for one beat, e.g. a camera beat (rendered near 1024 px) next to sharp frozen moments; slow (minutes per second of film) Constraints: 480 to 2160.
faces	Faces (optional)	null	restore: repair the faces in this beat's clip (GFPGAN), after it is generated. Off by default; or a mapping with strength, min_size, smooth

Used in: FilmScene.beats.

BeatFrame

A frame of a beat of a film: the first or last frame of what that beat renders (after its trim), or the frame at a time. Chain beats with it: a beat whose image is the last frame of the frozen moment before it starts exactly where that moment ends. The beat must come first in the film (or be rendered); build --only renders a frozen moment on the way, a generated beat that is not rendered yet stands in with its start image.

Field	Type	Default	Description
beat	string	required	The id of the beat
at	enum or number	"last"	first, last, or seconds into the beat's clip (after its trim) Allowed values: "first", "last".
pick	enum	"last"	With at: last, sharpest takes the sharpest of the last few frames instead of the very last one Allowed values: "last", "sharpest".
section	string (optional)	null	The film section holding the beat (found by the beat id when omitted)

Used in: ImageInput.beat_frame.

Brand

A brand as tokens: what theme.from reads from a web page, a logo or a brand.yaml, or what you write yourself. Missing colours are derived from the ones given with the presets' contrast rules.

Field	Type	Default	Description
mode	enum (optional)	null	light or dark (default: what the colours say) Allowed values: "light", "dark".
colors	map of string to string	{}	Brand colours as #hex: accent (the main brand colour) and any of accent2, bg, bg2, ink, accent_ink
fonts	map of string to string or FontFile	{}	Brand fonts by role (display, body, mono): a Google Fonts family or {file: ...} of the project's own
radius	string (optional)	null	Corner radius of buttons and cards, e.g. 6px
logo	string (optional)	null	The brand's logo (an image file in the project), shown in the nav when nav.logo is not set
source	string (optional)	null	Where these tokens were read from (written by studio brand)

Used in: Theme.from.

CameraKey

Field	Type	Default	Description
t	number	required	Seconds
at	list of 3 numbers	required	Camera position Constraints: exactly 3 items.
look	list of 3 numbers	required	Point the camera looks at Constraints: exactly 3 items.

Used in: Take.camera.

CameraMove

A keyed camera move on a generated (moving) beat: the same keys as a frozen moment's camera, but the people keep walking. The start image is given a depth map, the camera path is flown through it to make a guide video (the moving subjects softened so they stay free), and the video model generates the beat along that guide, so the camera does what the keys say instead of what the prompt hopes.

Field	Type	Default	Description
keys	list of FreezeKey	required	Camera keys over the beat: {t, box, hold} exactly as in a frozen moment (freeze.camera). The camera rests on the first box until its t and on the last after its t Constraints: at least 1 items.
ease	enum	"inOutSine"	How the camera travels between boxes (as freeze.ease; flow keeps moving through keys) Allowed values: "linear", "inOutSine", "inOutQuad", "inOutCubic", "inOutQuart", "smoother", "flow".
backend	enum	"auto"	What generates the beat along the guide: ltx (LTX-2.3 IC-LoRA depth control, the depth guide a take uses), wan (Wan 2.2 Fun Control, the guide as the control video) or uni3c (Wan Uni3C camera controlnet; needs a model file that is not installed by default). auto: the one measured to follow the keys best (see docs/ENGINE.md) Allowed values: "auto", "ltx", "wan", "uni3c".
subjects	string	"the people"	Who is free to move, as you would say it to a person (found by Florence-2, cut out by SAM 2.1). The guide fixes the camera and the static world and only hints at where these are. Empty: nothing is freed (the guide pins everything, people included)
soften	number	0.045	How far the subjects are softened in the guide, as a share of the frame width (a Gaussian blur of their own depth). 0 keeps their exact depth, which pins their silhouettes too (they only slide with the camera); larger frees them more but tells the model less about where they are Constraints: 0 to 0.3.
guide_strength	number	1.0	How strongly the guide steers the model (LTX: the IC-LoRA guide strength; Wan: Fun Control has no scale, so this is the Uni3C strength) Constraints: 0 to 2.
parallax	number	0.5	Depth effect of the guide's camera (as freeze.parallax): near things shift against far ones as the camera moves, so the model sees a real camera move Constraints: 0 to 2.
dolly	number (optional)	null	Extra push-in of near things (default follows parallax) Constraints: 0 to 3.
continues	boolean	false	Carry the previous beat's camera across the cut: this beat's camera starts where the previous camera beat ended, with the speed it ended at (no ease to a stop at the boundary), and it starts on that beat's last frame, graded to match
res	list of 2 integers (optional)	null	Size the video model runs at (the frames are scaled to beat_res; ltx: multiples of 64, wan/uni3c: multiples of 16). Default: ltx about 1024x576 (more pixels let the subjects drift in tests; beat_restore adds the detail back), wan ~1280x720 Constraints: exactly 2 items.
anchor	number	0.6	ltx: how firmly the pure-camera render of the start image (the picture flown along the keys) is pinned as keyframes along the way, once a second and at the end. It keeps what things look like and sharpens close-ups (in tests: Laplacian variance 48-63 against 29-48 without, 66 in the pure render) and tightens the camera (0.8% against 1.3% zoom error). 0 = only the first frame is pinned Constraints: 0 to 1.
match	number (optional)	null	How far the generated frames are pulled back toward the base plate's tone and grain (the start image of the first beat of the chain, flown along the same camera): a video model drifts in contrast, haze and grain from clip to clip, and a cut shows it. 0 = off. Default: 0.7 for a beat in a chain of continues beats, else off Constraints: 0 to 1.
seed	integer (optional)	null	Seed for the generation (default: the beat's motion_seed, then seed)

Used in: Beat.camera.

CameraPreset

A named camera move on a generated beat: camera: dolly-in, or camera: {move: orbit-left, amount: 0.8, at: [0.4, 0.5]}. Everything a keyed move takes (backend, subjects, soften, guide_strength, res, anchor, match, seed) applies; the camera path comes from the move instead of keys.

Field	Type	Default	Description
keys	list of FreezeKey	[]	Not used by a named move (its path is the move's): leave it out
ease	enum	"inOutSine"	How the camera travels between boxes (as freeze.ease; flow keeps moving through keys) Allowed values: "linear", "inOutSine", "inOutQuad", "inOutCubic", "inOutQuart", "smoother", "flow".
backend	enum	"auto"	What generates the beat along the guide: ltx (LTX-2.3 IC-LoRA depth control, the depth guide a take uses), wan (Wan 2.2 Fun Control, the guide as the control video) or uni3c (Wan Uni3C camera controlnet; needs a model file that is not installed by default). auto: the one measured to follow the keys best (see docs/ENGINE.md) Allowed values: "auto", "ltx", "wan", "uni3c".
subjects	string	"the people"	Who is free to move, as you would say it to a person (found by Florence-2, cut out by SAM 2.1). The guide fixes the camera and the static world and only hints at where these are. Empty: nothing is freed (the guide pins everything, people included)
soften	number	0.045	How far the subjects are softened in the guide, as a share of the frame width (a Gaussian blur of their own depth). 0 keeps their exact depth, which pins their silhouettes too (they only slide with the camera); larger frees them more but tells the model less about where they are Constraints: 0 to 0.3.
guide_strength	number	1.0	How strongly the guide steers the model (LTX: the IC-LoRA guide strength; Wan: Fun Control has no scale, so this is the Uni3C strength) Constraints: 0 to 2.
parallax	number	0.5	Depth effect of the guide's camera (as freeze.parallax): near things shift against far ones as the camera moves, so the model sees a real camera move Constraints: 0 to 2.
dolly	number (optional)	null	Extra push-in of near things (default follows parallax) Constraints: 0 to 3.
continues	boolean	false	Carry the previous beat's camera across the cut: this beat's camera starts where the previous camera beat ended, with the speed it ended at (no ease to a stop at the boundary), and it starts on that beat's last frame, graded to match
res	list of 2 integers (optional)	null	Size the video model runs at (the frames are scaled to beat_res; ltx: multiples of 64, wan/uni3c: multiples of 16). Default: ltx about 1024x576 (more pixels let the subjects drift in tests; beat_restore adds the detail back), wan ~1280x720 Constraints: exactly 2 items.
anchor	number (optional)	null	How firmly the pure-camera render is pinned along the way (as a keyed move's anchor). Default: the move's tested value, 0.6 for dollies, zooms, whip pans and static shots, 0.3 for moves that open gaps behind near things (orbit, crane, dolly zoom, fpv, push-through), which the model then repaints Constraints: 0 to 1.
match	number (optional)	null	How far the generated frames are pulled back toward the base plate's tone and grain (the start image of the first beat of the chain, flown along the same camera): a video model drifts in contrast, haze and grain from clip to clip, and a cut shows it. 0 = off. Default: 0.7 for a beat in a chain of continues beats, else off Constraints: 0 to 1.
seed	integer (optional)	null	Seed for the generation (default: the beat's motion_seed, then seed)
move	enum	required	A named camera move (engine/camera_moves.py), flown through the start image's depth and followed by the video model like keyed moves: static (locked-off tripod), dolly-in / dolly-out (the camera travels toward or away from at, near things grow faster than far ones), crash-zoom (a sudden lens zoom into at, then a hold), dolly-zoom (vertigo: the subject keeps its size while the background swells), orbit-left / orbit-right (an arc around the subject, which stays put while the foreground and background slide past each other), crane-up / crane-down (the camera rises or sinks, tilting to keep the subject), fpv (a drone flying forward with a banking roll), whip-pan (a fast pan with motion blur), push-through (a deep push past the foreground into the scene beyond). See docs/scenes.md for which read clearly on LTX-2.3 Allowed values: "static", "dolly-in", "dolly-out", "crash-zoom", "dolly-zoom", "orbit-left", "orbit-right", "crane-up", "crane-down", "fpv", "whip-pan", "push-through".
amount	number	1.0	How far the move goes: 1 is the tested recipe, 0.5 half as far, 1.5 further (a deeper push, a wider arc, a faster zoom) Constraints: 0.25 to 2.0.
at	list of 2 numbers (optional)	null	The point the move is about, x y in 0-1 of the start image (the subject): where a dolly or a zoom goes, what an orbit or a crane keeps still, what a dolly zoom keeps the same size. Default: the centre Constraints: exactly 2 items.

Used in: Beat.camera.

Carry

An element that travels across the cut: it leaves its place in this section and settles in its place in the next (FLIP: the browser measures both), so the eye is led over the join.

Field	Type	Default	Description
from	string	"h1, h2"	CSS selector of the element in this section (the last match on its final screen)
to	string	"h1, h2"	CSS selector of the element in the next section (the first match on its first screen)
via	enum	"dissolve"	What the rest of the picture does while the element travels Allowed values: "dissolve", "cut", "wipe", "iris".

Used in: Transition.carry.

Cinemagraph

Make a picture a cinemagraph: a still where only one thing moves (steam from a cup, a curtain in a breeze, light on water, a candle flame), and everything else is held perfectly still. Put it on any image input (hero.image, a product or strip image, a gallery tile, a section backdrop):

image: { generate: "a cup of coffee, steam rising ...", cinemagraph: { moves: "the steam above the cup" } }

The still stays the picture. The engine finds what moves (Florence-2 + SAM 2.1 from moves), has a local video model (LTX-2.3 or Wan 2.2) bring a crop around it to life, keeps only the moving part, closes the loop and encodes a small video of that part; the page lays it over the still through a soft mask, so outside the mask the pixels are the still's and never change. Plain images (no cinemagraph) are untouched and keep their cache.

Field	Type	Default	Description
moves	string	required	What moves, as you would say it to a person: "the steam above the cup", "the sheer curtains", "the light on the water", "the candle flame". Found on the still by Florence-2 grounding; it must be visible in the picture
motion	string (optional)	null	How it moves, for the video model: "slow, curling upward", "swaying a little in a soft breeze". Default: gentle motion of moves
seconds	number	4.0	Length of the loop in seconds (a pingpong loop is generated at half this and played forward then back) Constraints: 1.0 to 8.0.
loop	enum	"seamless"	seamless: the clip is made to end where it began (first = last frame generation) and any seam left is closed by a flow-aligned crossfade; the motion keeps one direction (steam keeps rising). pingpong: forward then backward, a seam by construction; for sways and shimmer that read the same both ways Allowed values: "seamless", "pingpong".
strength	number	1.0	How far it moves: 1 is the motion as generated, lower damps it toward its resting look, higher exaggerates it Constraints: 0.2 to 1.5.
seed	integer	7
model	enum (optional)	null	Video model: ltx (LTX-2.3) or wan (Wan 2.2). Default: the site's video_model, then ltx Allowed values: "ltx", "wan".
box	list of 4 numbers (optional)	null	x0, y0, x1, y1 around what moves, 0-1 of the still (skips the search by name) Constraints: exactly 4 items.
points	list of list of 2 numbers	[]	Points ON what moves, 0-1 across and down the still Constraints: each: exactly 2 items.
negative_points	list of list of 2 numbers	[]	Points on what must NOT move Constraints: each: exactly 2 items.
pick	enum or integer	"all"	When moves matches several things: all of them, or one by rank or index Allowed values: "all", "first", "largest", "smallest", "left", "right", "top", "bottom", "center".
grow	integer	10	Widen (+) or shrink (-) the moving region by this many pixels of the still (the video is free to move this much beyond it) Constraints: -60 to 200.
feather	integer	28	Softness of the region's edge in pixels of the still: the motion fades into the held-still picture over about this distance Constraints: 0 to 120.
track	boolean	true	Follow the moving thing through the clip (SAM 2.1 video tracking) and keep the whole path it sweeps, so a flame or a curtain that leaves its first outline is not cut off at the edge of the region

Used in: ImageInput.cinemagraph.

CodeFile

One file on the route through the codebase.

Field	Type	Default	Description
path	string	required	Shown as the file's title, e.g. app/Models/User.php
code	string (optional)	null	The source. For kind: edit, start each line with '+' (added), '-' (removed) or ' ' (unchanged), like a diff
file	string (optional)	null	Read the source from this project file instead of code
kind	enum	"read"	read: flies past as the agent reads it; edit: a stop where the diff applies itself as you scroll; run: a terminal stop whose output prints line by line Allowed values: "read", "edit", "run".
label	string (optional)	null	Tag over the file (default: Read / Edit / the command for run)
lang	string (optional)	null	Highlighting: php, ts, tsx, js, py, go, rust, sh... (default: from the extension)
weight	number	1.0	Share of the scroll this file gets, relative to the others of its kind
todo	string (optional)	null	An item on the agent's todo list, ticked when this file is done

Used in: CodeScene.files.

CodeTerminal

The agent's prompt box the scene opens on; the camera dives through it into the code.

Field	Type	Default	Description
logo	string	""	Wordmark over the prompt, e.g. 'opencode'; split as dim|bright with '|', e.g. 'open|code'
prompt	string	required
agent	string	"Build"
model	string	""
provider	string	""
hints	list of string	["tab switch agent", "ctrl+p commands"]

Used in: CodeScene.terminal.

CutAnchor

Where a shape sits on one side of a match cut: named (found on that side's picture by Florence-2 and SAM 2.1), boxed, or a page element. A bare string is a prompt: shape: "the round logo".

Field	Type	Default	Description
prompt	string (optional)	null	What to find on the picture, as you would say it to a person: "the sun", "the round logo", "the cup"
box	list of 4 numbers (optional)	null	x0, y0, x1, y1 around the shape, 0-1 of the section's picture Constraints: exactly 4 items.
selector	string (optional)	null	A CSS selector for a page element inside the section (the phone screen: .ss-phone__screen, a logo: .ss-wordmark svg); measured in the browser, so it works on any block

Used in: Transition.shape, Transition.becomes.

Device

A phone showing a simple app screen, drawn in HTML so the text stays sharp at any size.

Field	Type	Default	Description
app	string (optional)	null	App name in the screen's header
title	string (optional)	null	Screen title
label	string (optional)	null	Small label above the options
note	string (optional)	null	Paragraph under the label
options	list of DeviceOption	[]
button	string (optional)	null	Button at the bottom of the screen
image	string or ImageInput (optional)	null	Use a screenshot instead of the drawn screen
video	string (optional)	null	Play a screen recording (a project file or URL; muted, looping) instead of the drawn screen; image is then its poster
time	string	"9:41"

Used in: HeroBlock.device, ProductBlock.device, ProductBlock.behind.

DeviceOption

Field	Type	Default	Description
title	string	required
body	string (optional)	null
selected	boolean	false	Shown as the chosen option

Used in: Device.options.

DioramaCamera

How the camera moves: it orbits the whole diorama, dives to each place, and lifts back out.

Field	Type	Default	Description
orbit	enum	"auto"	Which way it travels round the world between places: the short way, or always clockwise / counter-clockwise (seen from above) Allowed values: "auto", "cw", "ccw".
pitch	number	36.0	Degrees above the horizon in the orbit view Constraints: 10 to 80.
dive_pitch	number	21.0	Degrees above the horizon once it has dived to a place Constraints: 6 to 60.
fov	number	30.0	Lens in degrees: narrow flattens the world into a macro photograph Constraints: 14 to 60.
dive	number	1.0	How close the dive lands (above 1 is closer) Constraints: 0.4 to 2.0.
hold	number	0.5	Share of each step's scroll the camera rests at its place; the rest is travel Constraints: 0.1 to 0.9.
drift	number	1.0	The camera's idle breathing, and how far the pointer can turn it (0: still) Constraints: 0 to 3.
travel	enum	"orbit"	Between two places: orbit lifts out over the whole world and dives to the next; glide goes straight from one to the next at the same height, following a scroll mover that travels between them, and at a place where that mover stops it looks at the mover and what it carries Allowed values: "orbit", "glide".
yaw	number (optional)	null	Degrees round the world the wide views look from (0: from the front, 90: from the right). Unset: beside the first and last places

Used in: DioramaScene.camera.

DioramaLook

The miniature look: shallow tilt-shift depth of field, high soft sun, toy saturation.

Field	Type	Default	Description
tilt_shift	number	1.0	Depth-of-field strength: 0 is sharp everywhere; 1 reads as a macro lens on a model; above 1 is extreme Constraints: 0 to 2.5.
saturation	number	1.15	Colour boost (toys are a little too colourful) Constraints: 0.5 to 2.0.
sun	enum	"golden"	Time of day: the sun's height and warmth, the length of the shadows Allowed values: "morning", "noon", "golden", "dusk".
table	string (optional)	null	Colour of the surface the diorama stands on (default: the theme background)
grain	number	0.03	Film grain Constraints: 0 to 0.2.
vignette	number	0.25	Darkening towards the corners Constraints: 0 to 1.0.

Used in: DioramaScene.look.

DioramaMover

Something that travels a route on the diorama: a vehicle, a cart, a carrier. With drive: scroll its place on the route follows the page: it waits at each stop while that stop's step is on screen and travels between stops as the page scrolls, and the camera follows it there (camera.travel: glide). With drive: time it goes round its route on its own, at speed.

Field	Type	Default	Description
id	string	required	A name for it
kind	string	"carrier"	What it is, a prop of the kit: carrier (a flat, levitating deck), forklift, boat ... (unknown kinds are drawn as a block)
route	list of list of 2 numbers	[]	The way it goes, points [x, z] in world units in order; corners are rounded. Its stops are joined to it in order Constraints: each: exactly 2 items.
stops	list of DioramaStop	[]	drive: scroll: where it waits and what it carries from there on, in route order
drive	enum	"scroll"	scroll: its place follows the page (see above); time: it loops its route on its own Allowed values: "scroll", "time".
speed	number	1.2	drive: time: world units per second Constraints: at most 20.
carry	string or ImageInput (optional)	null	What it carries at the start, a picture laid on its deck
carry_size	list of 2 numbers	[1.0, 0.9]	The carried picture's width and depth, world units Constraints: exactly 2 items.
scale	number	1.0	Constraints: at most 5.
color	string (optional)	null	Its main colour, hex
hover	number	0.0	How high it floats over the floor (a levitating deck), world units Constraints: 0 to 1.

Used in: DioramaScene.movers.

DioramaPlace

A place on the diorama: where the camera dives to and where one step's content sits. The place gets a hero prop, a sign with its name, a little set dressing around it and a pin in the orbit view.

Field	Type	Default	Description
name	string	required	The name on the sign and the pin
kind	string (optional)	null	The hero prop. harbour: lighthouse, crane, warehouse, stall, office, ferry, windmill, shop, cafe. town: clocktower, shop, cafe, windmill, stall, office, fountain. desk: laptop, mug, books, plant, lamp, monitor, notebook. factory: tank, chimney, conveyor, robot, warehouse, office, silo. studio: easel, softbox, camera, monitor, speaker, sofa, shelf, plant. Unset: the next one the world suggests
where	string or list of 2 numbers (optional)	null	Where on the world: coordinates [x, z] in world units from the middle, or words: a compass point (north is the back, south the front: 'north-east', 'south'), 'centre', 'edge', or one of the world's own spots (harbour: hill, yard, plaza, pier, bay, village, quay; the others in docs/miniature.md). Unset: the world's own spot for this kind Constraints: exactly 2 items.
step	integer (optional)	null	Which of the scene's steps (0-based) lives here. Unset: the places take the steps in order, skipping a leading intro step Constraints: at least 0.
side	enum (optional)	null	Which side of the screen the place sits on once the camera has dived to it (the copy takes the other). Unset: opposite the card column Allowed values: "left", "right".
sign	string (optional)	null	Text on the little sign at the place (default: the name; an empty text: no sign)
scale	number	1.0	Size of the hero prop Constraints: above 0.
color	string (optional)	null	Main colour of the hero prop, hex
turn	number (optional)	null	Degrees the hero prop is turned (0 faces the viewer, 90 faces right). Unset: the world decides (towards the water, or the viewer)

Used in: DioramaScene.places.

DioramaProp

One extra thing set on the diorama by hand, beside what the world generates.

Field	Type	Default	Description
kind	string	required	A prop of the kit: tree, pine, bush, rock, house, shop, cafe, warehouse, lamp, bench, fence, crate, barrel, flag, container, boat, windmill ... (the full vocabulary is in docs/miniature.md)
at	list of 2 numbers	required	Where, in world units from the middle (x to the right, z towards the viewer) Constraints: exactly 2 items.
turn	number	0.0	Degrees the front is turned (0 faces the viewer)
scale	number	1.0
color	string (optional)	null	Main colour (roof, hull, body), hex
size	list of 2 numbers (optional)	null	Width and depth in world units, for props that cover an area (tiles, rug) Constraints: exactly 2 items.

Used in: DioramaWorld.props.

DioramaStop

A stop on a mover's route: where it waits while a step is on screen, and what it carries from there on.

Field	Type	Default	Description
place	string (optional)	null	The name of a place: the mover stops at that place (its position on the route), while that place's step is on screen
at	list of 2 numbers (optional)	null	Or a point [x, z] in world units; it waits there while step is on screen Constraints: exactly 2 items.
step	integer (optional)	null	With at: which step (0-based) it waits through Constraints: at least 0.
carry	string or ImageInput (optional)	null	What it carries from this stop on (a picture laid on its deck); it changes while the mover waits here

Used in: DioramaMover.stops.

DioramaTraffic

Many movers wandering a grid on their own, each its own way: they drive straight along the grid's lines, stop, turn, and never run into each other or into what stands on the grid (the routes are worked out at build time, a few minutes of traffic that the page plays on a loop). Warehouse robots, planar-motor decks, carts in a yard.

Field	Type	Default	Description
area	list of 4 numbers	required	x0, z0, x1, z1: the rectangle they move in, world units Constraints: exactly 4 items.
cell	number	1.0	The grid's spacing, world units: one move is one cell Constraints: 0.3 to 6.
count	integer	12	How many movers Constraints: 1 to 80.
speed	number	1.6	World units per second while moving Constraints: at most 20.
pause	number	0.25	Seconds they rest at a cell before moving on Constraints: 0 to 5.
kind	string	"carrier"	What each is, a prop of the kit (see DioramaMover.kind)
carry	list of string or ImageInput	[]	Pictures laid on their decks; mover i carries picture i (mod the list), so a few pictures make a varied crowd
carry_size	list of 2 numbers	[0.7, 0.6]	The carried picture's width and depth, world units Constraints: exactly 2 items.
scale	number	1.0	Constraints: at most 5.
colors	list of string	[]	Hex colours for the movers, used in turn
hover	number	0.0	How high they float over the floor, world units Constraints: 0 to 1.
keep_off	list of list of 4 numbers	[]	More rectangles (x0, z0, x1, z1) they never enter; the routes of scroll movers and the props standing in the area are kept off by themselves Constraints: each: exactly 4 items.
seed	integer	7

Used in: DioramaScene.traffic.

DioramaWorld

The world, described once. The engine lays out the ground (water, roads, hills, paving), the scenery and the places' dressing from this and the seed; the same spec always gives the same world.

Field	Type	Default	Description
kind	enum	"harbour"	The kind of world: sets the ground plan, the scenery and the prop vocabulary. plain: an empty floor, for a world made of its places, props and movers Allowed values: "harbour", "town", "desk", "factory", "studio", "plain".
seed	integer	7	Seed for the layout: another seed is another arrangement of the same kind of world
size	number (optional)	null	Width of the diorama in world units (a house is about 2). Unset: 30 for a harbour or town, 22 for a desk or studio, 26 for a factory Constraints: 16 to 60.
base	enum (optional)	null	The shape of the base it stands on (default: round for harbour and town, square for the rest) Allowed values: "round", "square".
palette	list of string	[]	Hex colours for roofs, hulls and containers, replacing the world's own (2-8)
density	number	1.0	How much scenery fills the world: trees, houses, boats, lamps Constraints: 0 to 2.0.
props	list of DioramaProp	[]	Extra props set by hand
floor	string (optional)	null	plain: the floor's colour, hex (default: a pale concrete)

Used in: DioramaScene.world.

EditRegion

Where an image edit is allowed to change the picture; everywhere else the result is the source, pixel for pixel.

Name the region (prompt, found on the source image by Florence-2 grounding), or point at it (box, points; 0-1 picture coordinates). SAM 2.1 then cuts a precise mask around it. A bare string is a prompt: region: "the girl's shoes". The editor works on a padded crop around the region (see scope), and the result is composited back through the feathered mask at the source's own resolution.

Field	Type	Default	Description
prompt	string (optional)	null	What the edit may change, as you would say it to a person: "the girl's shoes", "the shoes of the woman and the man"
points	list of list of 2 numbers	[]	Points ON the thing, 0-1 across and down the source image (with or without a prompt or box) Constraints: each: exactly 2 items.
negative_points	list of list of 2 numbers	[]	Points on what the mask must NOT take Constraints: each: exactly 2 items.
box	list of 4 numbers (optional)	null	x0, y0, x1, y1 around it, 0-1 of the source (skips the prompt's search) Constraints: exactly 4 items.
pick	enum or integer	"all"	When the prompt matches several things: all of them, one by rank (largest, left, center ...) or by index (0 is the first the search lists) Allowed values: "all", "first", "largest", "smallest", "left", "right", "top", "bottom", "center".
within	string (optional)	null	Keep only the matches inside the thing this names, e.g. prompt "hands" within "the girl in the red coat". A prompt shaped "the X's Y" or "the Y of the X" does this by itself
grow	integer	0	Widen (+) or shrink (-) the mask by this many pixels of the source image (the edit is free to redraw this much beyond the thing itself) Constraints: -80 to 200.
feather	integer	8	Softness of the mask's edge in pixels of the source image: the edit fades into the untouched picture over about this distance Constraints: 0 to 100.
invert	boolean	false	Edit everything EXCEPT the region (the region stays the source, pixel for pixel)
scope	enum	"crop"	crop: the editor sees only a padded crop around the region, upscaled to its working size, so a small region gets full detail; image: the editor sees the whole picture (always used with invert) Allowed values: "crop", "image".
context	number	0.6	scope crop: how much of the surroundings the editor sees, as a share of the region's size added on every side Constraints: 0 to 4.

Used in: ImageEdit.region, JourneyStop.region.

Faces

Face repair for generated video: faces above min_size are found in every frame, restored with GFPGAN v1.4, and blended back softly (and smoothed over time so nothing flickers).

Field	Type	Default	Description
mode	"restore"	"restore"
strength	number	0.7	How much of the restored face is blended in Constraints: 0 to 1.
min_size	integer	96	Faces smaller than this many pixels across are left alone Constraints: 16 to 2048.
smooth	number	0.6	Temporal smoothing of the blend (higher: steadier, slower to follow) Constraints: 0 to 0.95.

Used in: Beat.faces, FilmScene.faces.

FaqItem

Field	Type	Default	Description
q	string	required
a	string	required

Used in: FaqBlock.items.

Feature

Field	Type	Default	Description
tag	string (optional)	null
title	string	required
body	string (optional)	null
meter	number (optional)	null	0-1 progress bar under the card
image	string or ImageInput (optional)	null	An illustration at the top of the card: a file, URL or image input (generate:, ...). In a named medium (imagery.medium) it is drawn in it like every generated picture, and shown whole

Used in: FeaturesBlock.items.

Finish

The finishing pass over a finished film (engine/finish.py), after it is generated and joined and before it is encoded for the page: a SeedVR2 restore (grain and compression mush out, fine detail redrawn), a deflicker and a light temporal denoise, an optional sharpen, and frame interpolation where the motion is slow. Each step is its own cached stage keyed by the film and its settings. finish: true takes every default.

Field	Type	Default	Description
restore	integer	1088	Run SeedVR2 at this height (1088 keeps a 1920x1088 film's size and redraws its detail; 1440 or 2160 enlarge it). null skips it. Slow: about a minute and a half per second of film at 1088 Constraints: 480 to 2160.
batch	integer	21	SeedVR2 frames restored together (4n+1). More frames: steadier detail from frame to frame (less shimmer), more VRAM Constraints: 5 to 49.
deflicker	boolean	false	Even out frame-to-frame brightness flicker (FFmpeg deflicker, over 5 frames). Off by default: on a moving camera it chases real changes of light and measured worse; turn it on for a clip that visibly pulses
denoise	number	0.25	Temporal denoise: pixels that only shimmer from frame to frame are averaged with their neighbours in time (FFmpeg atadenoise; edges and real motion are kept by its thresholds). 0 = off, 1 = strong (can smear fine moving texture) Constraints: 0 to 1.
sharpen	number	0.3	A light unsharp mask after the denoise (0 = off; 0.3, the default, gives back the fine detail the denoise softens without halos; higher shows halos) Constraints: 0 to 1.
interpolate	enum	"off"	Double the frame rate with a frame interpolator (smoother slow scrolling, twice the frames in the page's film). auto: only with a RIFE/FILM checkpoint in ComfyUI and only when the film's motion is slow (slow), where interpolation is clean; on: always, with FFmpeg's minterpolate when there is no checkpoint (it leaves halos along moving edges) Allowed values: "off", "auto", "on".
slow	number	6.0	interpolate auto: the most motion, in pixels a frame at 1920 wide (the film's median), that still counts as slow Constraints: at most 60.
interpolator	string (optional)	null	A RIFE or FILM checkpoint in ComfyUI's models/frame_interpolation folder (ComfyUI's own FrameInterpolate node). Default: the first one ComfyUI lists

Used in: FilmScene.finish.

FontFile

A font file of the project's own (a brand face that is not on Google Fonts), served from the site itself.

Field	Type	Default	Description
file	string	required	The font file inside the project folder: .woff2 (best), .woff, .ttf or .otf
family	string (optional)	null	The family name it is set in (default: from the file name, fonts/Brand-Sans.woff2 is 'Brand Sans')
weight	string (optional)	null	The weight the file holds, e.g. 400 or 700, or a range such as '100 900' for a variable font (default: every weight)
style	enum	"normal"	normal, or italic for an italic file Allowed values: "normal", "italic".

Used in: Brand.fonts, Theme.fonts.

FrameMatch

An image designed to match another section's last frame, so a match cut can be built rather than found: the shape that ends that section becomes something else in this picture, in the same place at the same size. edit hands the frame to Qwen-Image-Edit with an instruction made from these fields; hint generates this picture from prompt with the shape's place and size written into it. The section must come before this picture's in the page (it is built first); pair it with transition: {type: match} on that section.

Field	Type	Default	Description
section	string	required	The id of the section whose last frame to match
shape	string	required	What to match in that frame, as you would say it to a person: "the round logo"
becomes	string (optional)	null	What that shape is in this picture: "the sun", "the rim of a coffee cup"
prompt	string (optional)	null	The rest of this picture (the scene around the shape); with how: hint this is the whole prompt
how	enum	"edit"	edit: redraw that frame (Qwen-Image-Edit) keeping the shape's place, size and outline. hint: generate this picture (Z-Image) from prompt with the shape's place and size in it Allowed values: "edit", "hint".
at	enum	"last"	Match the section's last frame (or its first) Allowed values: "last", "first".
seed	integer	7
steps	integer	30	how: edit, the editor's steps Constraints: 4 to 60.

Used in: ImageInput.match.

Freeze

A frozen moment: the beat is not generated but rendered from one frame, offline, in seconds. The frame is upscaled, given a depth map (Depth Anything V2) and the camera moves through it as a 2.5D parallax move along camera keys, so a crowd can stand still while the camera visits a couple's hands and feet, and the faces stay the faces of the frame. The clip joins the film like any generated beat (trim, crossfade, restore, locked, --only).

Field	Type	Default	Description
from	string (optional)	null	Freeze the last frame of this beat (an earlier beat of the film, after its trim); or give image:
image	string or ImageInput (optional)	null	Freeze this image instead (a file, or any image input; beat_frame: takes a frame of a beat)
pick	enum	"sharpest"	With from: sharpest takes the sharpest of the last pick_window frames (the very last of a moving shot is often motion-blurred, bad for close-ups), last the very last one. Sharpness is measured in the camera's tightest boxes Allowed values: "sharpest", "last".
pick_window	integer	6	How many of the last frames pick: sharpest looks at Constraints: 1 to 48.
upscale	number	1.0	Upscale the frame this many times before moving through it, so crops stay sharp (a 3x crop of a 1920 frame needs 2). 1 = off Constraints: 1 to 4.
upscale_method	enum	"auto"	seedvr2: SeedVR2 image restoration through ComfyUI (redraws fine detail; the model downloads on first use). lanczos: a local detail-preserving resample, no model. auto: seedvr2 when its weights are installed, else lanczos Allowed values: "auto", "seedvr2", "lanczos".
camera	list of FreezeKey	[]	Camera keys over the beat. The camera rests on the first box until its t and on the last after its t; none: the whole frame, with the parallax
move	FreezeMove (optional)	null	A named camera move through the frozen frame instead of camera keys (see FreezeMove): move: dolly-zoom, or {move: orbit-left, amount: 0.8, at: [0.5, 0.6]}. The same rigs as a generated beat's named moves, rendered offline through the frame's depth: nothing in the picture moves but the camera
parallax	number	0.6	Strength of the 2.5D depth effect (0 = flat Ken Burns; the subject plane stays put while nearer and farther things shift) Constraints: 0 to 2.
dolly	number (optional)	null	Extra push-in of near things as the camera closes in (default follows parallax) Constraints: 0 to 3.
ease	enum	"inOutSine"	How the camera travels between boxes. inOutSine/Quad/Cubic/Quart: eased, resting at every key. smoother: eased with no jolt at the stops (it settles into each hold and leaves it gently; best with holds). flow: keeps moving through keys (it rests only at keys with a hold) Allowed values: "linear", "inOutSine", "inOutQuad", "inOutCubic", "inOutQuart", "smoother", "flow".
depth_model	enum	"small"	Depth Anything V2 size (small is Apache-2.0; base and large are CC-BY-NC) Allowed values: "small", "base", "large".
blur	number	0.0	Motion blur: the shutter, as a fraction of a frame (0 = off, 0.5 = 180 degrees). Sub-frames are averaged Constraints: 0 to 1.
grain	number	0.0	Film grain laid over the render (0.01-0.03 is faint) so it matches generated shots Constraints: 0 to 0.2.
seed	integer	7
detail	FreezeDetail (optional)	null	Deep zoom: a pyramid of re-drawn tiles so a push to a 10-20x close-up stays sharp at every scale (see FreezeDetail). detail: {} for the defaults
layers	FreezeLayers (optional)	null	Experimental. Layered depth: background plate + a plane per subject, so depth jumps have no halos and the camera reveals real background behind people (see FreezeLayers). layers: auto or a list of subjects

Used in: Beat.freeze.

FreezeDetail

Deep zoom: the camera may push from the wide shot to a 10-20x close-up and every scale stays sharp. The frame is not enlarged, it is built into a pyramid of tiles (about ratio times the detail per level, as many levels as the tightest camera box needs): each tile is a crop of its parent, enlarged (SeedVR2), re-drawn at low strength so it gains fine detail in the same style, aligned to its parent, and given the parent's low frequencies exactly, so colours and shapes agree at every level and only fine detail is new. The render takes each pixel from the finest tile that covers it and fades levels in by screen-space pixel density, so detail arrives as the camera does and nothing pops. Every tile is its own cached stage. Needs ComfyUI (SeedVR2, Z-Image Turbo).

Field	Type	Default	Description
levels	"auto" or integer	"auto"	Levels of tiles below the frame: auto stops when the tightest camera box has a tile pixel for every output pixel; or a number (1-8)
ratio	number	2.0	The most detail one level adds to the one above (pixels per unit of the frame). auto levels lower it so the levels share the gain evenly; a smaller ratio means more levels and tiles, and no moment where the picture is much softer than the screen Constraints: 1.3 to 4.0.
strength	number	0.35	How much each tile is re-drawn (img2img denoise): 0 keeps the enlargement as it is, 0.3-0.5 invents fine detail (stitching, creases, paper grain) without moving a shape Constraints: 0 to 0.9.
prompt	string	""	What the re-draw should add, in the picture's own words, e.g. 'hand-drawn ink, fine cross-hatching, stitched leather'. Added to a generic 'same style, fine detail' line
tile	number	1.25	A tile's resolution as a multiple of the beat's (1.25 = 2400 px across for a 1920 beat): more means fewer, sharper tiles and a slower build Constraints: 1.0 to 2.0.
lock	number	2.5	How firmly a tile agrees with its parent: the tile is made to match the parent at every scale coarser than this many parent pixels (colour, tone, where shapes are); finer than that it is its own. Lower holds the parent's soft edges and detail too, so a tile stays as soft as its parent; higher lets the tile move mid-sized shapes Constraints: 0.2 to 8.0.
method	enum	"zimage"	The re-draw: zimage (Z-Image Turbo img2img: keeps every shape where it is, fast; adds texture and creases, and draws what the prompt names where the picture already has the shape) or qwen (Qwen-Image-Edit asked to add fine detail: it understands the picture and draws laces and stitching, but also moves and reshapes things by several percent, which a tile cannot be made to agree with: ghosting. Experimental) Allowed values: "zimage", "qwen".
seed	integer (optional)	null	Seed of the re-draw (default: the freeze's seed)

Used in: Freeze.detail.

FreezeKey

One stop of a frozen moment's camera: the framing box the camera has reached at time t of the beat.

Field	Type	Default	Description
t	number	required	When the camera reaches the box, 0-1 of the beat Constraints: 0 to 1.
box	list of 4 numbers	required	Framing box [x0, y0, x1, y1] in 0-1 of the frozen frame. The aspect comes from beat_res: a box of another shape is widened (never cropped) to it, so everything in the box stays in frame Constraints: exactly 4 items.
hold	number	0.0	The camera rests on the box for this much of the beat after reaching it (the frozen moment still breathes with the parallax) Constraints: 0 to 1.

Used in: CameraMove.keys, CameraPreset.keys, Freeze.camera.

FreezeLayer

A subject that gets its own plane.

Field	Type	Default	Description
id	string (optional)	null
prompt	string (optional)	null	The subject as you would say it: 'the woman in the black coat', 'the people'. Every match is cut out (Florence-2 finds it, SAM 2.1 cuts it out)
box	list of 4 numbers (optional)	null	Or a box [x0, y0, x1, y1] around it, 0-1 of the frozen frame (skips the search) Constraints: exactly 4 items.
points	list of list of 2 numbers	[]	Points ON the subject, 0-1 Constraints: each: exactly 2 items.

Used in: FreezeLayers.subjects.

FreezeLayers

Layered depth: the picture is cut into planes, back to front: a background plate with the subjects painted out (what was really behind them), and a plane per subject (or group of subjects at a similar depth) with its own smooth depth and a soft alpha edge. The camera then reveals true background behind a person instead of smearing the person, and a depth jump no longer leaves a halo or a double edge. The planes are cut from the same picture (and, with detail, the same tiles) at render time, so every level of a deep zoom is layered too.

Field	Type	Default	Description
subjects	"auto" or list of FreezeLayer	"auto"	auto: every person found on the frame (Florence-2 'people'); or the subjects to cut out
planes	integer	3	At most this many subject planes: subjects at a similar depth share one Constraints: 1 to 8.
plate	string	""	What the empty background should look like where the subjects stood (added to 'remove the people, continue the background in the same style'), e.g. 'the empty foggy street'
grow	integer	3	How far (pixels of the frame at beat resolution) the hole in the background plate extends past a subject's edge Constraints: 0 to 60.
feather	number	1.5	Softness of a subject's alpha edge (pixels at beat resolution) Constraints: 0 to 12.
seed	integer (optional)	null	Seed of the background fill (default: the freeze's seed)

Used in: Freeze.layers.

FreezeMove

A named camera move through a frozen frame (freeze.move): the rig a generated beat's named move flies (engine/camera_moves.py), rendered offline through the frame's depth instead of guiding a video model. Nothing in the picture moves but the camera, so a dolly zoom or an orbit holds every face and every pose of the frame. Moves that open gaps behind near things show the frame's own pixels stretched there (2.5D: nothing behind a shape was ever drawn), so keep amount near 1 on a crowded picture.

Field	Type	Default	Description
move	enum	required	A named camera move (engine/camera_moves.py), flown through the start image's depth and followed by the video model like keyed moves: static (locked-off tripod), dolly-in / dolly-out (the camera travels toward or away from at, near things grow faster than far ones), crash-zoom (a sudden lens zoom into at, then a hold), dolly-zoom (vertigo: the subject keeps its size while the background swells), orbit-left / orbit-right (an arc around the subject, which stays put while the foreground and background slide past each other), crane-up / crane-down (the camera rises or sinks, tilting to keep the subject), fpv (a drone flying forward with a banking roll), whip-pan (a fast pan with motion blur), push-through (a deep push past the foreground into the scene beyond). See docs/scenes.md for which read clearly on LTX-2.3 Allowed values: "static", "dolly-in", "dolly-out", "crash-zoom", "dolly-zoom", "orbit-left", "orbit-right", "crane-up", "crane-down", "fpv", "whip-pan", "push-through".
amount	number	1.0	How far the move goes: 1 is the tested recipe, 0.5 half as far, 1.5 further Constraints: 0.25 to 2.0.
at	list of 2 numbers (optional)	null	The point the move is about, x y in 0-1 of the frame (the subject): where a dolly or a zoom goes, what an orbit or a crane keeps still, what a dolly zoom keeps the same size. Default: the centre Constraints: exactly 2 items.

Used in: Freeze.move.

GalleryItem

Field	Type	Default	Description
title	string	required
href	string	required
kicker	string (optional)	null
body	string (optional)	null
image	string or ImageInput (optional)	null	The tile's picture: a file or URL, or an image input (generate:, edit:, cinemagraph: ...)
video	string (optional)	null

Used in: GalleryBlock.items.

Glow

A soft radial light laid over a gradient.

Field	Type	Default	Description
x	number	0.5	Centre, 0-1 across the image
y	number	0.5	Centre, 0-1 down the image
color	string	required	Hex colour
radius	number	0.4	Softness, as a fraction of the longer side
strength	number	0.8	0-1: how far the centre goes to the glow's colour

Used in: Gradient.glows.

Gradient

A backdrop rendered locally with no model at all: a colour sweep, optional glows, stars and film grain. For colour cards, dusk skies and night seas behind type.

Field	Type	Default	Description
colors	list of string	required	Hex colours along the sweep, first to last Constraints: at least 2 items.
angle	number	180	Sweep direction in degrees, as in CSS: 180 is top to bottom, 90 left to right
glows	list of Glow	[]
grain	number	0.035	Film grain, 0 to 0.2
stars	integer	0	Faint stars scattered over the upper part, for night skies

Used in: ImageInput.gradient.

Hud

Overlay instruments tied to scroll progress.

Field	Type	Default	Description
kind	enum	"none"	Allowed values: "telemetry", "scale", "progress", "legend", "none".
fields	list of map	[]	telemetry: [{label, unit, from, to, curve, decimals, grouping}] values go from from (default 0) to to with progress^curve
events	list of map	[]	telemetry: [{at, label}] lit when progress >= at
clock	map (optional)	null	telemetry: {prefix, start, end, zero} seconds; zero = progress where T=0
scales	list of string	[]	scale: one label per level, e.g. ['100 µm','10 µm']
legend	list of map of string to string	[]	legend: [{label, color}]

Used in: SceneBase.hud.

ImageEdit

Derive an image from another with a local image editor (Qwen-Image-Edit): change one thing, keep the rest.

Field	Type	Default	Description
source	string or ImageInput	required	The image to edit (a file, or another generated image)
instruction	string	required	What to change, e.g. 'close the door completely, keep everything else'
refs	list of string or ImageInput	[]	Up to two more images the instruction can refer to ('put the second image on the laptop screen')
seed	integer	7
steps	integer	30	Constraints: 4 to 60.
region	EditRegion (optional)	null	Edit only inside this region of the source image (or, with invert, everywhere but inside it): the rest of the result is the source, pixel for pixel, so consecutive versions of a frame differ only where the edit is

Used in: ImageInput.edit.

ImageInput

An image the engine uses as input: a file in the project's inputs/, one generated locally, a gradient, or a still of the site's product.

Field	Type	Default	Description
file	string (optional)	null	Path relative to the project folder (inputs/photo.jpg) or an https URL
generate	string (optional)	null	Prompt for a local text-to-image model (Z-Image Turbo)
gradient	Gradient (optional)	null	Render a gradient backdrop (no model needed) at size
product	enum or ProductShot (optional)	null	A still of site.product at size: a view name (three-quarter, front, side, top, back, hero-low, exploded, detail) or a shot {view, part, yaw, pitch, zoom, explode, background, scene ...} Allowed values: "front", "three-quarter", "side", "top", "back", "hero-low", "exploded", "detail".
scene	string (optional)	null	With product: render it into a generated scene described by this prompt (see ProductShot.scene)
size	list of 2 integers	[1920, 1088]	Generated size (multiples of 16) Constraints: exactly 2 items.
seed	integer	7
crop	list of 4 numbers (optional)	null	Crop [x0, y0, x1, y1] in 0-1 image coordinates Constraints: exactly 4 items.
edit	ImageEdit (optional)	null	Make this image by editing another one locally (Qwen-Image-Edit)
beat_frame	BeatFrame (optional)	null	A frame of a beat of a film (its first or last frame, or the frame at a time): chain one beat onto the end of another
match	FrameMatch (optional)	null	Design this image to match another section's last frame, for a match cut (see FrameMatch)
credit	string (optional)	null	Attribution shown in the footer
cinemagraph	Cinemagraph (optional)	null	Make this picture a cinemagraph: only the thing named by moves moves (a video laid over the still through a soft mask), everything else is held perfectly still. Works wherever the page shows the image (hero, product, strip, gallery, a section backdrop)
medium	string (optional)	null	Draw this image in a named medium: risograph, toile, engraving, gouache, watercolour, halftone, ink (a tested recipe: the model's prompt plus a post-process in the theme's inks; see imagery.medium), over the site's own. Works on a file too (the post-process only). none leaves this image out of the site's named medium
calm	enum (optional)	null	Keep this side of the picture calm for a headline: asked of the model, then measured; a busy result is mirrored or regenerated with the next seeds (a file is only checked). Set by itself on a hero in a named medium (its align side) Allowed values: "left", "right", "center", "top", "bottom".
live	list of LiveDetail (optional)	null	Living details: one small thing alive in the still (see LiveDetail): sweep, glint, drift, rise or flow in a region, drawn by the page (no GPU needed). Up to four; on hero, product, strip and gallery images and parallax photos Constraints: at most 4 items.

Used in: ArtworkScene.image, Beat.image, Beat.end_image, Device.image, DioramaMover.carry, DioramaStop.carry, DioramaTraffic.carry, Feature.image and 18 more.

InsideLight

One key of an inside flight's light, at a point of the scene's progress. Unset fields follow the scene's own.

Field	Type	Default	Description
at	number	required	Scene progress, 0-1 Constraints: 0 to 1.
exposure	number (optional)	null	Exposure in stops Constraints: -6 to 4.
ambient	number (optional)	null	Strength of the studio light Constraints: 0 to 4.
lamp	number (optional)	null	Strength of the lamp that travels with the camera Constraints: 0 to 10.
rim	number (optional)	null	Strength of the light behind the stop's part Constraints: 0 to 10.
lamp_color	string (optional)	null	Colour of the camera lamp, hex

Used in: InsideScene.light.

InsideStop

One stop of the camera inside the product: it flies to a named part (or a point), lands there at macro scale with a shallow depth of field, and the stop's copy (the step in the same position) arrives with it. Only part (or point) is needed; everything else has a default that suits any model with named parts.

Field	Type	Default	Description
part	string (optional)	null	Glob on the part, mesh or assembly name, case-insensitive: 'driver*', '*magnet*', 'Housing/*' (see studio product <project> --parts). Several parts are framed together. Use point for a model with no named parts
point	list of 3 numbers (optional)	null	Instead of a part: a point inside the product, x y z in 0-1 of its bounds (x left to right, y front to back, z bottom to top) Constraints: exactly 3 items.
aim	list of 3 numbers	[0.5, 0.5, 0.5]	Where in the part's own bounds the camera looks, x y z in 0-1 (z 1 = its top surface): look at one end of a long part, or its top Constraints: exactly 3 items.
label	string (optional)	null	Name for this stop in studio inside and in alt text (default: the part's name)
span	number (optional)	null	How much fills the frame at the stop: the width of the view at the focal plane, in millimetres of the real product (macro: 8-40). Default: the part with a margin, never wider than 40% of the product Constraints: above 0.
lens	number (optional)	null	Focal length in mm (default: the scene's lens). Longer flattens and blurs more Constraints: above 4.
fstop	number (optional)	null	Aperture (default: the scene's fstop). 1.8 is a razor thin plane in focus, 8 keeps most of the part sharp Constraints: above 0.5.
focus	string (optional)	null	Glob on the part the focus is held on (default: the stop's part). Rack focus by naming a part in front of or behind it
yaw	number (optional)	null	Which side the camera comes from, degrees (0 the front, 90 the right side). Default: chosen per stop, the clearest way in, close to the previous stop's side
pitch	number (optional)	null	Camera height above the horizon, degrees (default: chosen, 15-40) Constraints: -85 to 89.
roll	number	0.0	Tilt of the horizon at the stop, degrees (a few degrees reads as handheld/cinematic)
explode	number (optional)	null	How far the parts are pulled apart at this stop (default: the scene's explode); opens a closed assembly so the camera can enter Constraints: 0 to 1.5.
hold	number	0.8	Scroll distance, in screens, the camera rests here while the copy is read Constraints: above 0.
travel	number	1.4	Scroll distance, in screens, of the flight that arrives at this stop Constraints: above 0.

Used in: InsideScene.stops.

Journey

The whole page as one camera journey through one generated world: each stop is a place in it, the camera travels between them in one unbroken move and the page's content arrives at its stop.

The engine builds it from what it already does best: the first picture is generated from the world and the first place; every later place is arrived at by a keyed camera move on a moving shot (flown through the picture before, painted by the video model), then frozen as a moment with a slow parallax while the page holds. Every join is on the same frame, so there is no cut, and every picture comes out of the one before, so there is no change of style. The runtime maps scroll to the film: a move plays as you scroll through it, a stop holds while its blocks are read, and the camera is at rest whenever the page is. On a phone the wide picture is cropped around focus; with reduced motion (or no JavaScript) the page is the stops as stills with their blocks, in order.

Field	Type	Default	Description
world	string	required	The world, described once: the place, its materials, the light and the hour ('a stone farmhouse on a northern moor at dawn, blue mist, oak and slate, firelight inside'). Added to the prompt of every picture and move, so the stops are one place. Say nothing about people
stops	list of JourneyStop	required	The places, in order Constraints: at least 2 items.
pace	number	1.0	Scales every scroll distance (holds and moves): 0.8 is brisker Constraints: at most 4.
rail	boolean	true	A rail of the places down the right edge: a mark per stop, the current one named, each a link that scrolls to it
seed	integer	11	Seed of the pictures and moves (each stop adds its position)
fade	number	0.25	Seconds of cross-dissolve between the move and the hold on the same frame (invisible; it only hides the one-frame seam of two encodes) Constraints: 0 to 1.
grain	number	0.012	Chain: film grain laid over the frozen holds so they match the generated moves on either side (as freeze.grain) Constraints: 0 to 0.2.
blockout	JourneyBlockout (optional)	null	The generation approach. Unset: a chain of camera moves between stop pictures (the camera flies through the picture the last stop left, so every stop must be visible in the one before: a facade, a street, a landscape). Set: one continuous take through a blockout of the world (real 3D, so doors and rooms work); every stop then needs a pose
restore	integer (optional)	null	Run SeedVR2 over the finished film at this height (e.g. 1080) to clean the video model's softness and redraw fine detail; slow Constraints: 360 to 2160.

Used in: FilmScene.journey.

JourneyBlockout

The world as rough shapes, as a take's: Blender renders their depth along the camera path, a still of every stop is painted over that depth (Z-Image Turbo + depth ControlNet) and the video model paints the world between the stops along the same depth (LTX-2.3). The camera is a real 3D path, so it can go through a door into a room the first picture never showed; the price is describing the world's layout once, in shapes.

Field	Type	Default	Description
objects	list of WorldObject	[]	Boxes, cylinders, blobs, planes, scatter: only layout, scale and depth matter, the models paint the detail. Leave a gap in a wall for a doorway (two boxes and a lintel)
ground	WorldObject (optional)	null	A plane under everything (default: 400 x 400 at z = 0)
sky	string	"#b9cde0"	Colour of the sky in the blockout
light	list of 3 numbers	[0.0, -0.4, 0.9]	Light direction of the blockout preview Constraints: exactly 3 items.
lens	number	28.0	Focal length in mm
depth_far	number	400.0	Depth guide far plane
control_strength	number	0.75	How tightly each stop's picture follows the blockout's depth (Z-Image depth ControlNet) Constraints: 0 to 1.
keyframe_res	list of 2 integers	[1536, 864]	Size of each stop's picture Constraints: exactly 2 items.
ltx_res	list of 2 integers	[1024, 576]	Size the video model runs at (multiples of 64); the film is scaled to the page. 1024x576 is the size it holds its look at Constraints: exactly 2 items.
guide_strength	number	1.0	How strongly the depth guide steers the video model Constraints: 0 to 2.
guide_blur	number	1.5	Blur of the depth guide in pixels: keeps the camera and the big shapes, drops the blockout's hard edges so a box does not stay a box Constraints: 0 to 12.
segment	number	8.0	Longest stretch the video model paints in one go, in seconds. The film is cut into stretches at the middle of each hold (the camera is at rest there, so the seam cannot show) and a longer one is split evenly Constraints: 3 to 10.
keyframe_strength	number	0.6	How firmly each stop's picture is pinned in the film. Too firm (0.9 was measured) and the model, holding the picture before and the picture after, dissolves between them instead of travelling; about 0.5-0.6 lets the depth guide move the camera and still lands on the picture Constraints: 0 to 1.
negative	string (optional)	null	What the video model must avoid (default: cuts, flicker, warped shapes, text, watermark, blur)

Used in: Journey.blockout.

JourneyMove

How the camera gets from the stop before to this one: a keyed camera move on a moving shot (camera: of a beat), flown through the picture the last stop left and painted by the video model, so the place it arrives at is the one this stop's place describes, found inside the last picture and not cut to.

Field	Type	Default	Description
into	list of 4 numbers (optional)	null	Chain (no blockout): the part of the previous stop's picture the camera travels into: [x0, y0, x1, y1] in 0-1 of that picture, the doorway, the window, the table. The camera arrives with this box filling the frame. Default: the middle third Constraints: exactly 4 items.
via	list of list of 4 numbers	[]	Chain: boxes the camera passes on the way (a bend in the path: past the fire and on to the stairs), evenly spaced in time Constraints: each: exactly 4 items.
path	list of JourneyPose	[]	With a blockout: poses the camera passes on the way (evenly spaced in time), a bend in its path: round the table, through the arch
seconds	number	4.0	Length of the move in the film (a chain: at most 10) Constraints: 1 to 20.
scroll	number (optional)	null	Scroll distance of the move, in screen heights (default: half the seconds of film). The page lands on the stop at the end of it Constraints: 0.3 to 12.
motion	string (optional)	null	What the move looks like, in words, for the video model: 'the camera glides through the open door into the hall, firelight ahead'
ease	enum	"inOutSine"	How the camera travels (as freeze.ease); the default starts and ends at rest, so the page lands softly at the stop Allowed values: "linear", "inOutSine", "inOutQuad", "inOutCubic", "inOutQuart", "smoother", "flow".
parallax	number	0.5	Depth effect of the guide (as camera.parallax) Constraints: 0 to 2.
anchor	number	0.6	How firmly the pure-camera render of the picture is pinned along the way (as camera.anchor): more keeps the world the picture shows, less lets the model paint more of the place it arrives at Constraints: 0 to 1.
guide_strength	number	1.0	How strongly the depth guide steers the model Constraints: 0 to 2.
res	list of 2 integers (optional)	null	Size the video model runs at (as camera.res) Constraints: exactly 2 items.
seed	integer (optional)	null	Seed of the move (default: the stop's). With a blockout it re-rolls the stretch of the take that ends at this stop (and so the stretches after it, which continue from it)

Used in: JourneyStop.move.

JourneyPose

Where the camera is and what it looks at, in the blockout's world units (metres, z up, y forward).

Field	Type	Default	Description
at	list of 3 numbers	required	Camera position Constraints: exactly 3 items.
look	list of 3 numbers	required	The point the camera looks at Constraints: exactly 3 items.

Used in: JourneyMove.path, JourneyStop.pose.

JourneyStop

One place on the journey. The camera arrives (move), rests while the page holds (hold), shows its blocks laid out to sit with the picture, and moves on.

Field	Type	Default	Description
id	string	required	Short name; the beats are called <id>-travel and <id>-hold
label	string (optional)	null	Name of the place in the page's places rail (default: the id)
world	string (optional)	null	Replaces journey.world in this stop's prompts: for a place the shared world would mislead, such as the view out of a window in a journey described from inside
place	string	required	The place as a photograph: what the camera sees at rest here, in the world described once at journey.world. The first stop's picture is made from it; later stops arrive at it through move (it steers the video model toward the place) and it names the picture for the reduced-motion page
still	string or ImageInput (optional)	null	Use this picture for the stop instead (a file, or an image input: generate:, edit:, beat_frame ...). For the first stop it replaces the picture made from place; for a later one the hold freezes it instead of the frame the camera arrived at
edit	string (optional)	null	Make the stop's picture by editing the frame the camera arrived at (Qwen-Image-Edit): 'a wooden crate of jars stands on the table'. With region the rest of the frame is that frame, pixel for pixel, so the join stays invisible
region	EditRegion (optional)	null	edit: where it may change the picture
pose	JourneyPose (optional)	null	With a blockout: where the camera rests at this stop
move	JourneyMove	{} (its own defaults)	How the camera gets here (not used by the first stop)
frame	list of 4 numbers (optional)	null	The framing the hold settles on, a box in 0-1 of the stop's picture (default: a slow push-in of breathe). The hold starts and ends on the whole picture, so it joins the moves before and after exactly Constraints: exactly 4 items.
breathe	number	0.05	How far the camera pushes in during the hold when frame is not set, as a share of the picture. The frozen moment keeps its parallax, so it lives Constraints: 0 to 0.3.
seconds	number	3.0	Length of the hold in the film (chain: a frozen moment, no model, rendered offline in seconds; blockout: the camera at rest) Constraints: 1 to 20.
hold	number	1.4	Scroll distance of the hold, in screen heights: how long the page stays at this place while its blocks are read Constraints: 0.3 to 12.
parallax	number	0.6	Depth effect of the hold (as freeze.parallax) Constraints: 0 to 2.
focus	list of 2 numbers (optional)	null	The point of the picture (0-1 across, 0-1 down) a portrait phone keeps in view when it crops the wide picture (default: the middle) Constraints: exactly 2 items.
blocks	list of one of 10 models, chosen by type: IntroBlock, FeaturesBlock, StatsBlock, TimelineBlock, QuoteBlock, CtaBlock, GalleryBlock, HeroBlock, ProductBlock, FaqBlock	[]	Content shown at this stop (hero, intro, features, stats, quote, faq, product, cta ...): the same blocks as a section, set as part of the place instead of over it
side	enum	"auto"	Where the blocks sit on the picture. auto: the calmest part of the picture, found from the picture itself (least detail, so nothing important is covered), alternating sides Allowed values: "auto", "left", "right", "center", "top", "bottom".
form	enum	"auto"	How the blocks are set. type: bare lettering with a soft scrim, as if written in the scene. rail: items in a column divided by hairlines, like a plan beside the place. cards: panels of glass. plaque: one framed plate with a rule inside, like an engraved sign. auto: each block type picks the one that suits it Allowed values: "auto", "type", "rail", "cards", "plaque".
surface	enum	"auto"	light/dark: the picture under the blocks is light (dark lettering) or dark (light lettering). auto: measured from the picture where the blocks sit Allowed values: "auto", "light", "dark".
seed	integer (optional)	null	Seed of this stop's picture and move (default: the journey's, plus the stop's position)

Used in: Journey.stops.

Level

Field	Type	Default	Description
preset	string	required	Level generator, see engine/scenes/scene3d.py PRESETS
params	map	{}

Used in: Scene3DScene.levels.

LightLamps

Lights that switch on (or off) between two times: windows, street lamps, a lit cabin. Name them and the build cuts them out of the lit picture (Florence-2 + SAM 2.1, engine/region.py); the page then turns each one on at its own moment as the light changes, with a glow, instead of fading the whole picture.

Field	Type	Default	Description
prompt	string	required	What turns on, as you would say it to a person: 'the street lamps and the lit windows'. Found on the picture where it is lit
color	string (optional)	null	Glow colour, hex (default: measured from the lit pixels)
glow	number	1.0	Strength of the halo around a light as it comes on Constraints: 0 to 3.
grow	integer	2	Widen (+) or shrink (-) each cut-out by this many pixels Constraints: -20 to 60.

Used in: LightScene.lamps.

LightStructure

Geometry lock. The place must not move between times, only the light. Every relight is measured against the world (how far the editor displaced structure, and how well edges agree); its global move is undone; a relight that is still too far off is redone with another seed, and the best attempt is kept. What drift is left between neighbouring times is measured too and the page glides along it (a flow texture) instead of cross-fading doubled edges.

Field	Type	Default	Description
align	enum	"affine"	affine: undo the editor's global move (a few percent of zoom and a few pixels of shift, which is most of what it does to the place) and bend nothing; flow: also pull local structure onto the world with dense optical flow (edges agree better, lines wobble where the editor redrew things; for small drift); none: measure only. What drift remains is glided over in the page Allowed values: "affine", "flow", "none".
max_drift	number	8.0	How far the place may still be from the world after the global move is undone, mean displacement in pixels at 1024 across; above it the relight is retried with another seed Constraints: 0 to 40.
retries	integer	1	How many more seeds to try when a relight drifts too far Constraints: 0 to 8.
strength	number	0.88	How far the editor may depart from the picture it relights (the denoise strength of the edit): 1 starts from noise and relights fully but redraws and moves things; lower starts from the picture's own layout and holds it (0.75 moves things about 1 px, 0.88 a few) while relighting less. The prompts are written for 0.85-0.9 Constraints: 0.4 to 1.0.
chain	boolean	false	Relight each time from its neighbour nearer the world's time instead of from the world: smaller steps of light hold the layout and make neighbouring times agree better, at the price of a little sharpness per step

Used in: LightScene.structure.

LightSun

A sun (by day) and a moon (by night) that travel over the sky as the scroll goes through the day: a soft glow drawn in the browser over the sky only, behind the buildings. The relit pictures are asked for a sky without a sun disc so the two never double up.

Field	Type	Default	Description
rise	list of 2 numbers	[0.1, 0.5]	Where it comes up, 0-1 across and down the picture Constraints: exactly 2 items.
noon	list of 2 numbers	[0.5, 0.06]	Where it is highest Constraints: exactly 2 items.
set	list of 2 numbers	[0.92, 0.5]	Where it goes down Constraints: exactly 2 items.
size	number	1.0	Size of the glow Constraints: 0.1 to 4.
strength	number	1.0	Brightness of the glow Constraints: 0 to 3.
moon	boolean	true	A cool glow takes its place after dark, crossing the same way
sky	string (optional)	null	What the sky is called for the cut-out (default: 'the sky')

Used in: LightScene.sun.

LightTime

One time of day of a light scene. A bare string is a named preset (dawn, morning, noon, afternoon, golden-hour, sunset, blue-hour, night, overcast, storm; engine/light_presets.py) or, when it is not one, free text describing the light.

Field	Type	Default	Description
id	string (optional)	null	Name to pin steps to (state: {time: <id>}); default: the preset's name, else the first words of the prompt
preset	string (optional)	null	A named time of day: dawn, sunrise, morning, noon, afternoon, golden-hour, sunset, blue-hour, night, overcast, storm. It supplies the light's wording, the clock, the mist and whether lamps are on
prompt	string (optional)	null	Free text describing this light ('low green storm light, the sun breaking through on the left, wet stone'). With a preset it is added to the preset's wording
clock	string (optional)	null	The hour shown beside the picture, e.g. '18:40' (default: the preset's)
haze	number (optional)	null	Mist lying in the distance at this time, 0-1; the page draws it with the depth map and lets it drift (default: the preset's) Constraints: 0 to 1.
lit	boolean (optional)	null	Lamps and windows are lit at this time (default: the preset's). The build finds what turned on or off between neighbouring times
seed	integer (optional)	null	Seed for this time's relight (default: the scene's); try another when one drifts or looks wrong
strength	number (optional)	null	This time's own denoise strength (default: the scene's structure.strength): a time far from the world's light, such as night, needs more freedom (0.95) than one near it Constraints: 0.4 to 1.0.
image	string or ImageInput (optional)	null	This time's own picture of the same place, instead of relighting the world (a real photograph, a hand-fixed render). It is aligned to the world like any relight

Used in: LightScene.times.

Link

Field	Type	Default	Description
label	string	required
href	string	"#"
icon	enum (optional)	null	An icon on the button: arrow (after the label), play (before it), external (after it; the link opens in a new tab and says so to screen readers) or none. Default: the primary button's arrow, as before Allowed values: "arrow", "play", "external", "none".
look	enum (optional)	null	solid: filled with the accent; ghost: transparent with a thin outline, for a secondary action. Default: the first button of a row is solid Allowed values: "solid", "ghost".

Used in: CtaBlock.button, Footer.links, HeroBlock.buttons, IntroBlock.button, Nav.links, Nav.cta, ProductBlock.buttons, Step.button.

LiveDetail

One small thing alive in a still picture: a light sweeping, a cloud drifting, sparks rising, a glint, water moving. Drawn by the page itself (a small WebGL effect over the picture, or inside a parallax photo's own pass), so it needs no GPU at build time and works on a laptop. Paused off screen and in a hidden tab; with reduced motion the picture is simply still. Put it on a hero, product, strip or gallery image, or on a parallax photo:

image: { file: inputs/hero.jpg, live: [{ kind: drift, box: [0.35, 0.25, 0.55, 0.42] }] }

sweep: a soft band of light slides across the region now and then (or, with origin, a beam turns round that point). glint: small four-pointed twinkles on the region's brightest points (found in the picture at build time). drift: the region slides slowly to and fro with soft edges (clouds, mist, a hanging sign). rise: soft specks rise through the region and fade (embers, bubbles, dust in a beam). flow: a gentle wavy shimmer (water).

Field	Type	Default	Description
kind	enum	required	What lives there: sweep, glint, drift, rise or flow Allowed values: "sweep", "glint", "drift", "rise", "flow".
box	list of 4 numbers (optional)	null	x0, y0, x1, y1 of the region, 0-1 of the picture. For drift give some room around the thing that moves (it slides inside the box and fades out at its edges) Constraints: exactly 4 items.
prompt	string (optional)	null	Or name the region, as you would to a person ("the clouds"): found at build time by Florence-2 + SAM 2.1 (the same tooling as edit.region; needs ComfyUI). Without them the build uses box if given, and leaves the detail out otherwise
pick	enum or integer	"all"	When prompt matches several things: all of them, or one by rank or index Allowed values: "all", "first", "largest", "smallest", "left", "right", "top", "bottom", "center".
speed	number	1.0	How fast it lives: 1 is calm, 0.5 slower, 2 twice as fast Constraints: 0.1 to 4.
strength	number	0.6	How much: the brightness of a sweep, glint or rise, how far a drift slides, how much a flow ripples Constraints: 0.05 to 1.
color	string (optional)	null	The light's colour (sweep, glint, rise): a theme token (light, dark, accent, accent2, ink, bg, bg2) or a hex colour. Default: light (the lighter of the theme's text and background) for sweep and glint, accent2 for rise
angle	number	0	Direction in degrees, 0 to the right, 90 down: where a sweep slides and a drift goes; the line a flow's ripples run along. A rise always goes up Constraints: -360 to 360.
origin	list of 2 numbers (optional)	null	sweep: turn a beam round this point (x, y, 0-1 of the region, may lie outside it: a lighthouse lamp) instead of sliding a band Constraints: exactly 2 items.
count	integer	12	glint: how many twinkles; rise: how dense the specks are (about this many on screen) Constraints: 1 to 24.
softness	number	0.25	How soft the region's edge is, as a share of the box Constraints: 0 to 0.5.

Used in: ImageInput.live.

Logo

Field	Type	Default	Description
name	string	required	The company's name: the image's alternative text, or set as a wordmark when there is no image
image	string (optional)	null	The logo: an SVG or a transparent PNG/WebP in the project (or an https URL). Shown as a single-colour silhouette in the theme's text colour; a picture without transparency is shown in greys
href	string (optional)	null

Used in: LogoRow.items, LogosBlock.items.

LogoRow

A strip of partner or customer logos, in one colour so they sit quietly together.

Field	Type	Default	Description
title	string (optional)	null	A small line above the logos, e.g. 'Trusted by teams at'
items	list of Logo	required	Constraints: at least 1 items.
tint	enum	"mono"	mono: every logo in the theme's text colour (silhouettes; greys for pictures without transparency); original: their own colours Allowed values: "mono", "original".

Used in: HeroBlock.logos.

Marker

Field	Type	Default	Description
lng	number	required
lat	number	required
label	string	required

Used in: MapScene.markers.

MotionTokens

How the page moves. Each value defaults to what the runtime has always used; scroll.speed still paces it.

Field	Type	Default	Description
ease	string (optional)	null	Easing of transitions (default cubic-bezier(.22, .8, .2, 1))
enter	number (optional)	null	Share of the screen height a block scrolls through while it arrives (default 0.2) Constraints: at most 1.
follow	number (optional)	null	How quickly smooth scrolling catches up with the wheel (default 0.1; 0.5 is nearly immediate) Constraints: at most 0.5.

Used in: Theme.motion.

OrbitItem

Field	Type	Default	Description
label	string	required
status	string (optional)	null	e.g. Live, Developing, Roadmap; the first status used is highlighted

Used in: OrbitBlock.items.

PalettePool

only/avoid take palette names (studio directions --list palettes) or colour words (avoid: [red, green]).

Field	Type	Default	Description
only	list of string	[]	Consider only these (empty: all)
avoid	list of string	[]	Never use these
brand	list of string	[]	Hex colours every direction must carry as its accent: the palettes keep their grounds and take these

Used in: Directions.palettes.

PartMaterial

A material override for the product's parts whose names match. studio product <project> --parts lists the part, mesh and material names. Rules apply in order, so a later rule wins where two match.

Field	Type	Default	Description
match	string	"*"	Glob on the part, mesh or assembly name, case-insensitive: 'Case*', '*knob*', 'Housing/*' (an assembly name matches every part inside it)
material	string (optional)	null	Only the faces using a material whose name matches this glob (as listed by --parts), e.g. 'mat_2' or '*Rubber*'
color	string (optional)	null	Base colour, hex (sRGB)
metallic	number (optional)	null	0 plastic/paint, 1 bare metal Constraints: 0 to 1.
roughness	number (optional)	null	0 mirror, 0.3 satin, 0.6 bead-blasted, 0.9 matte Constraints: 0 to 1.
emission	string (optional)	null	Glow colour, hex (screens, LEDs)
emission_strength	number	1.0	Glow brightness when emission is set Constraints: at least 0.
transmission	number (optional)	null	See-through amount: 1 glass or clear acrylic Constraints: 0 to 1.
texture	string (optional)	null	Base-colour image (project file or URL). Meshes without UVs get a box projection
texture_scale	number	0.05	Box projection: metres per repeat of the texture Constraints: above 0.
region	list of 6 numbers (optional)	null	Only faces whose centre lies in this box, x0 y0 z0 x1 y1 z1 in 0-1 of the product's bounds (x left to right seen from the front, y front to back, z bottom to top): paint a screen window or a button on a one-piece body Constraints: exactly 6 items.
facing	enum (optional)	null	Only faces pointing this way (seen from the front view) Allowed values: "front", "back", "left", "right", "top", "bottom".

Used in: Product.parts.

Pointer

The visitor's pointer reveals the picture. A lens is a soft circle under the pointer that shows a second layer through the first: the picture's own colour through a black-and-white (or dimmed, blurred, tinted) page, or a second picture of the same view (before and after, the inside of a product through its outside). light is a torch: the picture is dark and a pool of warm light follows the pointer; on a photo with a depth map the light falls off with depth and rakes across the shapes. On a phone the lens follows the finger while it touches, a little above it so the finger does not hide it, and rests at rest otherwise, drifting gently. With reduced motion (and for keyboard users) it stays at rest. Purely visual: the words never depend on it, and without WebGL2 the picture shows as it always did. Works on parallax and film sections. A torch is drawn in 2D from the picture and its depth map, not relit from a 3D render.

Field	Type	Default	Description
mode	enum	"lens"	lens: a circle that shows layer through the picture. light: a torch, a pool of light on a darkened picture Allowed values: "lens", "light".
layer	"colour" or string or ImageInput or list of string or ImageInput	"colour"	lens: what shows through it. colour: the picture's own colour, with the rest in the base look. Or a second picture of the same view, the same shape as the photo (any image input: a file, generate:, edit:, a product still), or a list, one per photo of a parallax section in step order; a photo without one has no lens. A film's lens shows its colour
base	string (optional)	null	What the picture looks like outside the lens: mono (black and white), dim, blur, a hex colour (a duotone through it), or none (as it is). Default: mono when the lens shows colour (a film's own reveal.base if it has one), none when it shows a second picture
radius	number (optional)	null	Size of the lens or the pool of light: its radius as a share of the picture's shorter side. Default 0.2 for a lens, 0.32 for light Constraints: 0.04 to 0.6.
softness	number	0.35	How soft the edge is: 0 a sharp circle, 1 a wide fade Constraints: 0 to 1.
strength	number (optional)	null	lens: how fully the layer shows inside it (default 1). light: how dark the rest of the picture is, 0 not at all to 1 nearly black (default 0.85) Constraints: 0 to 1.
color	string	"#ffd7a8"	light: the colour of the light (a warm torch by default), a hex colour
rest	list of 2 numbers	[0.62, 0.42]	Where the lens or light rests when no pointer is on the picture (phones, keyboard, reduced motion): x, y in 0-1 of the picture Constraints: exactly 2 items.

Used in: FilmScene.pointer, ParallaxScene.pointer, StagesScene.pointer.

Pool

What a direction may draw on: only these (empty: everything), never these.

Field	Type	Default	Description
only	list of string	[]	Consider only these (empty: all)
avoid	list of string	[]	Never use these

Used in: Directions.styles, Directions.fonts.

ProductLayer

The real thing in a take: rendered from its own 3D model through the take's camera and composited over the generated world, so it never warps. One of model or from_photo, or neither (product: {}) to use the site's product: (its normalised model, materials and credit).

Field	Type	Default	Description
model	string (optional)	null	GLB/glTF, OBJ, FBX, STL or PLY: a project file or an https URL. Omit to use site.product
from_photo	string (optional)	null	A cut-out PNG (transparent background) turned into a 3D model with TRELLIS.2 (needs trellis_python/trellis_script in studio.toml)
credit	string (optional)	null	Attribution shown in the footer
at	list of 3 numbers	[0, 0, 0]	Where the base centre of the product sits in the world Constraints: exactly 3 items.
height	number (optional)	null	Height of the product in world units (default: site.product's real height in metres, else 1)
rotation	list of 3 numbers	[0, 0, 0]	Euler degrees Constraints: exactly 3 items.
spin	number	0.0	Extra yaw, in degrees, turned evenly over the whole take
hdri	string (optional)	null	Light for the product. Default: the plate itself (an environment built from the generated frames around each moment, so the product takes the scene's light and reflections). Or a Blender studio light: studio, forest, city, courtyard, night, interior, sunrise, sunset
sun_strength	number	3.0	A sun lamp along the take's light direction; 0 to turn it off
samples	integer	64	Cycles samples per frame (denoised)
shadow	boolean	true	Cast the product's shadow onto the generated ground
shadow_size	number (optional)	null	Side of the square the shadow falls on, in metres (the table or floor patch); default 3x the height
res	list of 2 integers (optional)	null	Render size of the product pass, which is also the final film size (default 1920x1080; 2560x1440 when the take is upscaled) Constraints: exactly 2 items.
fstop	number	2.8	Depth of field: focused where the camera looks, at this f-stop (real scale); none for everything sharp
motion_blur	number	0.5	Shutter as a fraction of a frame (0: off) Constraints: 0 to 1.
grade	boolean	true	Match the product pass to the plate: black level, white balance, saturation, softness, and a light wrap at the edges
grain	number (optional)	null	Film grain on the product (0-255 sigma); default: measured from the plate Constraints: at least 0.

Used in: Take.product.

ProductShot

A still of the site's product, rendered by Cycles in a studio light and cached.

Field	Type	Default	Description
view	enum	"three-quarter"	front, three-quarter, side, top, back, hero-low (a low camera looking up), exploded (parts pulled apart) or detail (a close macro view of part) Allowed values: "front", "three-quarter", "side", "top", "back", "hero-low", "exploded", "detail".
part	string (optional)	null	detail: glob on the part/mesh name to frame, e.g. '*port*' (see studio product --parts); also aims any other view at that part
yaw	number (optional)	null	Override the view's turn, degrees (0 front, 90 right side)
pitch	number (optional)	null	Override the view's camera height, degrees above the horizon
zoom	number	1.0	1 fits the product in the frame with a margin; 1.3 is closer Constraints: above 0.
explode	number (optional)	null	How far parts pull apart (exploded view: 1) Constraints: 0 to 1.5.
lens	number (optional)	null	Focal length in mm (default 50; hero-low 32)
background	string	"auto"	auto (transparent where the page shows it, such as hero and product blocks, else the theme background), transparent, theme, or a hex colour
hdri	string	"studio"	Blender studio light: studio, city, courtyard, forest, interior, night, sunrise, sunset
samples	integer	128	Cycles samples (denoised) Constraints: 4 to 4096.
shadow	boolean	true	A soft contact shadow under the product
exposure	number	0.0	Exposure in stops (-0.7 tames a pale metal lid in a bright scene) Constraints: -4 to 4.
fstop	number (optional)	null	Depth of field, focused on the product (or part); detail views default to f/8, a macro look Constraints: above 0.
scene	string (optional)	null	Render the product into a generated scene instead of the studio: a prompt for the place it stands ('a walnut desk by a window, morning light'). Z-Image paints the scene to the product's camera (perspective-matched), the product is lit by it, and a light Qwen-Image-Edit pass blends it in (needs ComfyUI)
harmonise	number	0.3	scene: strength of the blending pass (0 skips it); the product's own pixels are put back if the pass changes them Constraints: 0 to 1.
seed	integer	7	scene: seed for the generated scene

Used in: ImageInput.product.

Rating

A rating line: stars, the score and where it comes from ('4.9 from 860 reviews').

Field	Type	Default	Description
score	number	required	The score, e.g. 4.9 Constraints: at least 0.
out_of	integer	5	The best score; also the number of stars Constraints: 1 to 10.
text	string (optional)	null	Where the score comes from, e.g. 'from 860 reviews'
href	string (optional)	null	Link the line to the reviews
stars	boolean	true	Draw the stars (filled to the score)

Used in: HeroBlock.rating, SignupBlock.rating.

Reveal

Colour that comes in where the camera looks, driven by scroll. The film is rendered once in full colour; the page shows everything in the base look, and each region's colour is brought in through a tracked mask, in the browser (a WebGL shader draws each frame of the film with its masks from the same frame). Without WebGL the colour film plays as it is; with reduced motion each region jumps to colour as its moment arrives.

Field	Type	Default	Description
base	string	"mono"	What pixels that are not revealed look like: mono (black and white), dim (darker and muted), blur (soft focus in colour), or a hex colour such as "#2a4a6a" (a duotone from black through that colour to white)
regions	list of RevealRegion	required	Constraints: at least 1 items.
keys	list of map of string to number	[]	Scroll progress of the section -> how revealed each region is (0-1), eased between keys: [{at: 0.4, her_hand: 0}, {at: 0.48, her_hand: 1}]. A region keeps its last value. Default: one region after another
cuts	list of number	[]	Times of the film's cuts, in seconds, when they can't be known (a beats film knows its own; other films are searched for them)
feather	number	6.0	Softness of every mask's edge, in pixels at film size Constraints: 0 to 64.
ease	enum	"smooth"	How a region eases between two keys Allowed values: "smooth", "linear".
tracker	enum	"large"	SAM 2.1 size: large (sharper, ~450 MB) or base_plus Allowed values: "large", "base_plus".
strength	number	1.0	1: revealed pixels show the film's own colour; lower mixes the base look back in Constraints: 0 to 1.

Used in: FilmScene.reveal.

RevealAt

The frame a region is first found on (before it is tracked through the whole film): a moment in a beat (beat + t, 0 = its first frame, 1 = its last), or in the whole film (t alone, 0-1), or an exact frame / seconds of the joined film. Pick a frame where the thing is plainly in view.

Field	Type	Default	Description
beat	string (optional)	null	beats film: the beat's id
t	number	0.5	0-1 through the beat (or through the film without beat) Constraints: 0 to 1.
frame	integer (optional)	null	An exact frame of the film (overrides beat/t) Constraints: at least 0.
seconds	number (optional)	null	An exact time in the film, seconds (overrides beat/t) Constraints: at least 0.

Used in: RevealRegion.at, WorldPlace.anchor.

RevealRegion

One part of the picture that comes into colour. Describe it (prompt: found on the at frame by Florence-2), or point at it (points / box in 0-1 picture coordinates on the at frame); SAM 2.1 then follows it through the whole film, forwards and backwards.

Field	Type	Default	Description
id	string	required
prompt	string (optional)	null	What to colour, as you would say it to a person: "the woman's hand", "the shoes of the woman and the man". Every match on the at frame is revealed (see pick, within)
points	list of list of 2 numbers	[]	Points ON the thing, 0-1 across and down the at frame (with or without a prompt or box) Constraints: each: exactly 2 items.
negative_points	list of list of 2 numbers	[]	Points on what the mask must NOT take Constraints: each: exactly 2 items.
box	list of 4 numbers (optional)	null	x0, y0, x1, y1 around it, 0-1 on the at frame (skips the prompt's search) Constraints: exactly 4 items.
at	RevealAt or list of RevealAt	{} (its own defaults)	The frame to find it on. A film of several shots tracks a region only inside a shot it is found in (a cut ends it): give one at per shot it must show in
pick	enum or integer	"all"	When the prompt matches several things: all of them, one by rank (largest, left, ...) or by index (0 is the first the search lists) Allowed values: "all", "first", "largest", "smallest", "left", "right", "top", "bottom".
within	string (optional)	null	Keep only the matches inside the thing this names, e.g. prompt "hand" within "the woman in the long coat". A prompt shaped "the X's Y" or "the Y of the X" does this by itself
grow	integer	0	Widen (+) or shrink (-) the finished mask by this many pixels at film size Constraints: -40 to 80.

Used in: Reveal.regions.

Route

Field	Type	Default	Description
id	string	required
points	list of list of 2 numbers	required	[lng, lat] waypoints Constraints: each: exactly 2 items.
color	string (optional)	null
geodesic	boolean	false	Bend each leg along the great circle (flights)

Used in: MapScene.routes.

ShadeTokens

The shade laid over a scene's picture so its captions stay readable.

Field	Type	Default	Description
strength	number (optional)	null	How strongly the shade fades the picture toward the background: 1 as designed (the default), 0 none Constraints: 0 to 1.

Used in: Theme.shade.

SignupForm

An inline email field and a button: a plain HTML form that posts to the address you name (a form service, a newsletter's signup URL) or opens an email with mailto:. No server is involved; the browser checks the address.

Field	Type	Default	Description
action	string	required	Where the form goes: an https URL that accepts a form post, or mailto:you@example.com
method	enum	"post"	How the form is sent (a mailto: form always posts) Allowed values: "post", "get".
name	string	"email"	The field name the service expects Constraints: matches ^[A-Za-z][\w.\-\[\]]{0,63}$.
label	string	"Email address"	The field's label, read by screen readers (and shown when the placeholder is empty)
placeholder	string	"you@example.com"	Faint example text in the empty field
button	string	"Sign up"	The button's words
note	string (optional)	null	Small print under the field, e.g. 'No spam. Leave any time.'
hidden	map of string to string	{}	Extra hidden fields the service needs (a list id, a redirect URL)

Used in: HeroBlock.signup.

SpaceTokens

Page spacing. Each value defaults to what the stylesheet has always used.

Field	Type	Default	Description
gutter	string (optional)	null	Margin at the page's sides (default clamp(20px, 4vw, 64px))
max	string (optional)	null	Widest the content runs (default 1600px)
caption	string (optional)	null	Width of the captions over a scene (default min(680px, 82vw))

Used in: Theme.space.

SplatKey

Field	Type	Default	Description
t	number	required	Scene progress 0-1
at	list of 3 numbers	required	Camera position, in the scan's own coordinates Constraints: exactly 3 items.
look	list of 3 numbers	required	Point the camera looks at Constraints: exactly 3 items.

Used in: SplatScene.keys.

Stage

One state of the subject: a picture of exactly the same view as the others (same framing, same size).

Field	Type	Default	Description
id	string (optional)	null	Name to pin steps to (state: {stage: <id>}); default s0, s1 ...
image	string or ImageInput	required	The picture of this state; every stage is the same view
reveal	StageReveal	{} (its own defaults)	How this stage arrives over the one before (fade, wipe or map; ignored on the first)

Used in: StagesScene.stages.

StageFrame

How the pictures sit in the stage: cropped to fill it (cover) or whole (contain).

Field	Type	Default	Description
fit	enum	"cover"	cover: fill the stage, cropping what does not fit; contain: show the whole picture, the stage's background around it Allowed values: "cover", "contain".
position	list of 2 numbers	[0.5, 0.5]	contain: where the picture sits in the stage, 0-1 across and down ([1, 0.5] is against the right edge); cover: the point of the picture kept in view Constraints: exactly 2 items.
zoom	number	1.0	Scale on top of the fit (1.1 is a tenth larger) Constraints: 0.5 to 3.0.

Used in: StagesScene.fit, StagesScene.phone.

StageFront

A band of light that travels with a reveal's front: a scanning head, a band of heat, a sweep of light. It is drawn where the front is, so it moves exactly with the scroll.

Field	Type	Default	Description
color	string	"accent"	Its colour: a theme role (accent, accent2, ink, bg, bg2, light, dark) or a hex colour
width	number	0.02	How wide the band is, as a share of the front's travel Constraints: 0.002 to 0.3.
strength	number	0.8	How bright the band is Constraints: 0 to 2.
where	enum	"change"	change: the band shows only where the picture changes between the two stages (each part lights up as it lands); all: across the whole picture (a band of heat, a scan line) Allowed values: "change", "all".

Used in: StageReveal.front.

StageReveal

How a stage arrives over the one before. A bare string is the kind (fade, wipe, map).

Field	Type	Default	Description
kind	enum	"fade"	fade: the whole picture at once; wipe: a straight front crosses the picture along angle; map: each pixel arrives when the scroll reaches its grey in map (black first, white last), so parts can land one by one in any order Allowed values: "fade", "wipe", "map".
angle	number	0.0	wipe: the direction the front travels, in degrees (0 left to right, 90 top to bottom, 180 right to left)
map	string or ImageInput (optional)	null	map: a greyscale picture of the same view: when each pixel arrives (black first). Flat greys make whole parts arrive together
softness	number	0.06	How soft the front is, as a share of its travel (0 a hard edge) Constraints: 0.0 to 1.0.
front	StageFront (optional)	null	A band of light that travels with the front (see StageFront)

Used in: Stage.reveal.

Stat

Field	Type	Default	Description
value	string	""	Shown as-is; a plain number counts up. Leave empty when data fills it
unit	string (optional)	null
label	string	required
data	StatData (optional)	null	Read the value from a chart's data at build time, so headline numbers can never drift from the source

Used in: HeroBlock.stats, ProductBlock.stats, StatsBlock.items.

StatData

Field	Type	Default	Description
section	string	required	id of a chart section on this page
series	string	required
at	enum or number	"last"	Allowed values: "max", "min", "first", "last".
decimals	integer	0
scale	number	1.0	Multiply the value (e.g. 0.001 to show thousands)

Used in: Stat.data.

Step

One beat of copy that scrolls past the scene: a caption, or any block (block:).

Field	Type	Default	Description
kicker	string (optional)	null
title	string	""	Headline; <em>…</em> marks the accent words. Every caption step has one
body	string (optional)	null
fact	string (optional)	null	Small print under the body (mono)
at	list of 2 numbers or enum or AtBox or AtGrid (optional)	null	Where the step's words sit. A named spot: left, right, center, top, bottom, top-left, top-right, bottom-left, bottom-right (auto, the default, lets the layout decide). Or a box on the stage, {x, y, w, h, align} in 0-1 (x and y are its top-left corner). Or a slot on the theme's 12-column grid, {grid: "9 / span 4", row: top \| middle \| bottom}. Overlay scenes take all three; cards take all three for the card; split takes a spot (the copy stays in its column). Two numbers, [from, to], are the older way to write show Allowed values: "auto", "left", "right", "center", "top", "bottom", "top-left", "top-right", "bottom-left", "bottom-right".
length	number	1.2	split/cards layouts: scroll distance of this step, in screen heights
intro	boolean	false	Render as the section's opening headline (h1)
hint	string (optional)	null	Small 'scroll to begin' line under an intro step
place	enum or WorldPlace (optional)	null	overlay layout: where this caption sits on the shot (default: left, centred vertically); a corner here is the same as at: <corner>. Or an object, {kind: behind \| surface \| sky, ...}: the title is set inside the scene instead of over it (type in the world; see WorldPlace). place: sky is shorthand Allowed values: "left", "right", "center", "top-left", "top-right", "bottom-left", "bottom-right".
surface	enum (optional)	null	This caption's colours, when the shot under it is lighter or darker than the rest of the scene (dark words on a bright sky: light), with a soft glow behind them Allowed values: "light", "dark".
button	Link (optional)	null	A button under the caption, e.g. the call to action on a film's last shot
badge	string or Badge (optional)	null	A small pill above the headline: words, or {text, href, tag, icon} for a link with a 'New' tag or a dot
image	string or ImageInput (optional)	null	parallax: this step's photo (a step without one is another caption over the photo before it)
state	map	{}	Scene-specific state for this step. chart: {show, focus, x, y, reveal, marks}; map: {center, zoom, pitch, bearing, route, marker}
show	list of 2 numbers (optional)	null	When the step is on screen: from and to, 0-1 of the scene's progress (0.3-0.46 is the stretch from 30% to 46% of the way through). Overlay scenes; in split and cards a step shows while it scrolls past, for its length. Scenes that time their own copy (inside, light, code with state.file, film beats with state.beat, diorama) fill it in for steps that leave it out Constraints: exactly 2 items.
enter	enum (optional)	null	How the step arrives: fade, rise (the words rise into place), wipe (revealed from the left), type-on (word by word), scale (grows in) or none. Default: the layout's own, words rising over a stage and cards brightening as they reach the middle of the screen. With reduced motion only the fade is kept Allowed values: "fade", "rise", "wipe", "type-on", "scale", "none".
exit	enum (optional)	null	How the step leaves: fade, rise (the words rise away), wipe (cleared to the right), type-on (gone at once), scale or none. Default: the layout's own Allowed values: "fade", "rise", "wipe", "type-on", "scale", "none".
block	enum	"caption"	What the step shows: caption (the default: kicker, title, body, fact, hint, button) or a block, with that block's own fields beside block: (features, stats, timeline, quote, cta, gallery, product, faq). A block step is placed with at and timed with show like any other step; a style that draws the block its own way draws it that way here too Allowed values: "caption", "features", "stats", "timeline", "quote", "cta", "gallery", "product", "faq".

Used in: SceneBase.steps.

StripItem

Field	Type	Default	Description
image	string or ImageInput	required
caption	string (optional)	null

Used in: StripBlock.items.

Subject

The one subject a story film is about (a person, a product, an animal), kept the same from shot to shot. Every generated beat without its own image starts on a picture made from this reference by the local image editor (Qwen-Image-Edit): the beat's prompt as the new scene, the subject kept as it is in the reference. Within a shot the video model starts on that picture, so the subject keeps its identity across the film.

Field	Type	Default	Description
image	string or ImageInput	required	The reference: a file (inputs/hero.png) or an image input (generate:, a product still ...). A clear, well-lit picture of the subject alone works best
name	string	"the subject"	What the subject is, as you would say it to a person: 'the woman in the red coat', 'the white ceramic cup'. The editor is told to keep exactly that
keep	string	"face, hair, clothes, shape and colours"	What must not change from the reference
steps	integer	30	Image editor steps per start picture Constraints: 4 to 60.

Used in: FilmScene.subject.

Take

One continuous camera move through a generated world, with a real product locked to the same camera.

The world is a rough blockout (shapes + camera keys) rendered by Blender into a depth guide; LTX-2.3 paints the world along that guide in chained segments; the product is rendered by Cycles through the identical camera and composited over every frame; the result is encoded for scrubbing.

Field	Type	Default	Description
duration	number	required	Seconds
fps	integer	24
sky	string	"#8fb4d8"
light	list of 3 numbers	[0.4, -0.3, 0.85]	Light direction for the preview and the product's sun Constraints: exactly 3 items.
ground	WorldObject (optional)	null	A plane under everything (defaults to a 400 x 400 plane at z = 0)
objects	list of WorldObject	[]
camera	list of CameraKey	required	Constraints: at least 2 items.
lens	number	32.0	Focal length in mm
product	ProductLayer (optional)	null
product_in_guide	boolean	false	Put the product's shape in the depth guide. Off by default: the video model then paints its own version of the product, which shows around the edges of the real one. Leave the product out of the prompts too and describe the empty spot ('an empty table').
preview_res	list of 2 integers	[1280, 720]	Blockout render size; same aspect as product.res Constraints: exactly 2 items.
depth_far	number	400.0	Depth guide far plane
keyframes	map (optional)	null	Z-Image stills pinned at times, as in shot.keyframes
ltx	map	required	LTX-2.3 settings, as in shot.ltx: res, segments [{id, range, prompt}], seed, guide_strength ... Render at res [1920, 1088] (multiples of 32) for a detailed world; [1024, 576] is a quick draft
upscale	map (optional)	null	Optional upscale stage, as in shot.upscale
restore	integer (optional)	null	Run SeedVR2 over the generated world at this height (e.g. 1080) to clean the video model's softness and redraw fine detail, before the product is composited (the real product is never restored); slow Constraints: 360 to 2160.
video_model	enum (optional)	null	Video model: ltx (LTX-2.3, the default) or wan (Wan 2.2 A14B: renders at ~1280x720 and 16 fps, retimed to 24; clean, physical motion and first/last-frame control, but no finer detail than LTX after restore, and slower). Unset = the site's video_model, then studio.toml's, then ltx. With wan the world is painted by Wan 2.2 Fun Control along the same depth guide, from the first keyframe, in the segments and prompts of ltx: Allowed values: "ltx", "wan".
wan	Wan (optional)	null	Wan settings for this take (over the site's wan:)
locked	boolean	false	Keep the take's renders (world, product pass, composite) as they are, even when its settings change: build and storyboard log it and move on

Used in: FilmScene.take.

TimelineItem

Field	Type	Default	Description
when	string	required
title	string	required
body	string (optional)	null

Used in: TimelineBlock.items.

TrailerShot

One shot of the launch trailer, cut from something the site already rendered. Give one source: section (with step or range), beat, product or card.

Field	Type	Default	Description
section	string (optional)	null	id of a section. A film plays its own footage (take, beats or video); any other section is recorded scrolling through the built site (at phone size for 9:16)
step	integer (optional)	null	film: the footage under this step (its index in the section's steps), captioned with its headline
beat	string (optional)	null	id of a beat of a beats film (the section is found for you)
range	list of 2 numbers (optional)	null	Part of the source: seconds of a film, or 0-1 scroll progress of a recorded section Constraints: exactly 2 items.
product	string (optional)	null	A still of site.product from this view (front, three-quarter, side, top, back, hero-low, exploded) with a slow push-in. Needs the product feature; skipped with a note when it is missing
card	string (optional)	null	A type card on the theme background with this headline (<em>…</em> marks accent words)
caption	string (optional)	null	Headline over the shot (<em>…</em> marks accent words). Default: the matching step or section headline; '' for none
kicker	string (optional)	null	Small label above the caption
callouts	list of string	[]	Short feature lines ('24 h cold') set big, one after another on the beat, over this shot
beats	integer (optional)	null	Length on the trailer's beat grid (default: long enough to read the caption, at least a bar) Constraints: 1 to 64.
speed	number	1.0	Playback speed of the footage Constraints: 0.25 to 4.0.
hold	boolean	false	Freeze the first frame of range instead of playing it
zoom	list of 2 numbers (optional)	null	Push-in: scale at the start and the end of the shot (1 = the whole frame). Default: a slow push on the product reveal and on stills Constraints: exactly 2 items.
focus	list of 2 numbers (optional)	null	Point (0-1 x, y of the source) that the 9:16 crop and the push-in keep in frame. Default: the product in a take, else the most salient area Constraints: exactly 2 items.

Used in: Trailer.shots.

Transition

How a section hands over to the next, in the language of film. Without it sections simply follow one another (today's behaviour). The cut takes its own stretch of scroll (length): both pictures are held on screen while the scroll drives the cut, then the page carries on. Reduced motion, and pages without scripts, get a plain cut.

cut: an instant cut at a scroll position with a held beat either side. dissolve: a held cross-dissolve. wipe: the next picture is revealed along a line (direction, angle, soft). iris: it opens from a point as a circle (origin, or closing for the iris that closes on this one). whip: a fast directional blur-pan into the next (direction, blur). match: the end of this section and the start of the next share a shape; the frames scale and move until the two shapes coincide and the picture swaps on the shape (shape, becomes). carry: an element travels from its place here to its place there (carry).

Field	Type	Default	Description
type	enum	required	cut, dissolve, wipe, iris, whip, match or carry Allowed values: "cut", "dissolve", "wipe", "iris", "whip", "match", "carry".
length	number (optional)	null	Scroll distance the cut takes, in screen heights (defaults: cut 0.8, dissolve 1.2, wipe 1.3, iris 1.4, whip 1.1, match 1.5, carry 1.5) Constraints: 0.1 to 4.
hold	number	0.15	The share of the stretch held still at each end, so each picture is on screen a moment before and after the cut Constraints: 0 to 0.4.
at	number	0.5	cut and match: where in the cut (between the two holds) the picture swaps, 0-1 Constraints: 0.1 to 0.9.
direction	enum (optional)	null	wipe and whip: the way things travel (left: the next picture arrives from the right). Default left Allowed values: "left", "right", "up", "down".
angle	number (optional)	null	wipe: the edge's direction in degrees (0 right, 90 down); overrides direction
soft	number	0.0	wipe and iris: how soft the edge is (0 hard) Constraints: 0 to 1.
origin	list of 2 numbers (optional)	null	iris: where it opens from, 0-1 across and down the screen (default the middle; with a match, the shape) Constraints: exactly 2 items.
closing	boolean	false	iris: close on this section (the next shows behind it) instead of opening the next
blur	number	1.0	whip: blur strength, 1 is natural Constraints: 0 to 3.
shape	CutAnchor (optional)	null	match: the shape on this section's last picture. Leave out both shape and becomes to let the engine look for the best-matching pair of shapes (contours)
becomes	CutAnchor (optional)	null	match: the shape on the next section's first picture (default: the same prompt as shape)
zoom	number	3.0	match: the most each side may be scaled to bring the shapes to the same size (more than this and they only nearly coincide) Constraints: 1 to 8.
blend	number	0.0	match: a quick cross-dissolve over the swap, as a share of the stretch (0: a hard swap on the shape) Constraints: 0 to 0.3.
carry	Carry (optional)	null	carry: the element that travels (the section's heading to the next one's by default)

Used in: SceneBase.transition, BlockBase.transition.

TypeTokens

Type sizes. Each value defaults to what the stylesheet has always used.

Field	Type	Default	Description
scale	number (optional)	null	Multiplies every headline size, on phones too (default 1: 1.15 is a seventh larger) Constraints: at most 4.
body	string (optional)	null	Size of running text (default 17px)

Used in: Theme.type.

WorldObject

One rough shape in a take's world. Only layout, scale and depth matter: the video model paints the detail.

Field	Type	Default	Description
kind	enum	required	Allowed values: "box", "cylinder", "cone", "blob", "plane", "scatter".
name	string (optional)	null
at	list of 3 numbers	[0, 0, 0]	Centre (x, y, z); z is up Constraints: exactly 3 items.
size	list of 3 numbers	[1, 1, 1]	box/plane: full width, depth, height; blob: radii Constraints: exactly 3 items.
radius	number	1.0	cylinder/cone radius
depth	number	1.0	cylinder/cone height
rotation	list of 3 numbers	[0, 0, 0]	Euler degrees Constraints: exactly 3 items.
color	string	"#9a9a9a"
soft	boolean	false	blob: visible in the preview but left out of the depth guide (clouds, haze)
count	integer	40	scatter: how many boxes
seed	integer	1	scatter: random seed
area	list of 4 numbers	[-50, -50, 50, 50]	scatter: x0, y0, x1, y1 Constraints: exactly 4 items.
height	list of 2 numbers	[2, 10]	scatter: min and max box height Constraints: exactly 2 items.
footprint	list of 2 numbers	[3, 6]	scatter: min and max box width/depth Constraints: exactly 2 items.
shape	enum	"box"	scatter: box (buildings, rocks) or cone (pines) Allowed values: "box", "cone".
clear	list of 3 numbers (optional)	null	scatter: keep (x, y, radius) empty Constraints: exactly 3 items.

Used in: JourneyBlockout.objects, JourneyBlockout.ground, Take.ground, Take.objects.

WorldPlace

A headline that lives inside the scene. The camera moves and the type goes with the world: it is occluded by what is nearer, it sits on a surface in perspective, it hangs in the haze. The step's title is the text (its <em> words take the theme's accent as everywhere); kicker, body and the button stay an ordinary caption beside it. Works on a film or a sequence with layout: overlay and an at window on the step (a sequence: the product's own alpha is the silhouette, so type set behind it is cut exactly), and on a parallax section's step with a photo. Without WebGL, with reduced motion, and where the headline would be cropped off a narrow screen (a phone held upright), the title falls back to a clean overlay caption.

Field	Type	Default	Description
kind	enum	required	behind: set at a depth in the scene, so what is nearer (a pine, a railing, a building, a person) passes in front of it as the camera moves; surface: painted on a wall, a sign or a floor, tracked to it in perspective, with the surface's light and texture showing through; sky: large type in the haze at far depth, with fog in front Allowed values: "behind", "surface", "sky".
at	number or string (optional)	null	behind: what the type sits just behind, as a prompt ("the spruce", "the railing"; found on the anchor frame by Florence-2 + SAM 2.1, so it needs ComfyUI the first time) or a depth, 0 far to 1 near, in the shot's own range (default: behind the nearest third of what is in its box). surface: what it is painted on, as a prompt ("the wooden deck", "the white wall"; required unless quad is given). sky: a depth (default 0.07, just in front of the farthest thing)
x	number (optional)	null	Centre across, 0-1 of the frame (behind, sky) or of the surface (surface). Default 0.5
y	number (optional)	null	Centre down, 0-1 of the frame or of the surface. Default 0.5 (sky: 0.3)
size	number (optional)	null	Width of the type, as a fraction of the frame (behind, sky) or of the surface (surface). Default 0.5 behind, 0.72 sky, 0.8 surface Constraints: at most 1.5.
lines	integer (optional)	null	Break the title into this many lines (default: the layout that fills the width best) Constraints: 1 to 5.
align	enum	"center"	Alignment of a title of several lines Allowed values: "left", "center", "right".
caps	boolean	false	Set the title in capitals
track	number (optional)	null	Letter-spacing in em (default: 0 behind, 0.05 on a surface, 0.1 in the sky) Constraints: -0.1 to 0.6.
weight	integer (optional)	null	Font weight (default: the theme's display weight) Constraints: 100 to 900.
color	string	"auto"	auto (light or dark ink, whichever reads on the footage under it), ink, paper, accent, or a hex colour
anchor	RevealAt (optional)	null	The frame the thing is found on and the type is laid out on (default: the middle of the step's at window). Pick a frame where it is plainly in view
gap	number	0.05	behind (named thing): how far behind it the type sits, in depth units Constraints: 0 to 0.5.
follow	number (optional)	null	How much of the camera's motion the type follows (behind, sky): 1 moves with the layer it sits in, 0 stays on the screen. Default: what its depth says Constraints: 0 to 1.5.
haze	number (optional)	null	How much of the scene's air is mixed into the type (behind: 0.12, sky: 0.5): colour and contrast of distance Constraints: 0 to 1.
fog	number	0.55	sky: fog in front of the type, thicker low in the picture and where something far passes in front of it Constraints: 0 to 1.
blur	number (optional)	null	Softness of the type in pixels of a 1920-wide frame (behind, sky: default matches how sharp the footage is at that depth) Constraints: 0 to 30.
wear	number	0.35	surface: how much of the paint has worn off (the surface's own texture breaks the letters up) Constraints: 0 to 1.
relief	number	0.8	surface: how much of the surface's grain and light shows through the paint Constraints: 0 to 1.
quad	list of list of 2 numbers (optional)	null	surface: the four corners of the surface on the anchor frame, 0-1, clockwise from the top left, instead of finding it by prompt (the type is then tracked from these corners) Constraints: exactly 4 items; each: exactly 2 items.
angle	number	0.0	surface: turn the type on the surface, degrees Constraints: -180 to 180.
slant	number	1.0	surface: how much of the slant the depth map gives the surface is used (1: as estimated, 0: the type lies flat to the screen). The estimate is a guess from relative depth; lower it when the letters look over-foreshortened Constraints: 0 to 1.5.
track_model	enum (optional)	null	How the footage is followed: a surface is tracked as a plane seen in perspective (homography); behind and sky as a layer that shifts, turns and zooms (similarity) Allowed values: "homography", "similarity".

Used in: Step.place.