
The picture
CozyClay is a browser-based previs tool, described in its own repository as 'open source previs software in the browser: block a scene, pose characters, author camera moves and cuts, then take the same shots to an AI video model.' It is released under the AGPL-3.0-or-later licence and launched with 'npx cozyclay', opening a local studio on the user's own machine rather than a hosted service. As retrieved 16 September 2026, the project's own site states its purpose: 'Block the scene, pose the characters, set the camera, then hand that exact shot to an AI video model.'
What the documents show
The repository documentation is specific about what gets exported. Its Export menu produces 'a keyframe pack (first/last frame, clip, camera JSON, prompt, README) as a zip, an mp4 of the shot, depth + normal conditioning passes, a storyboard contact sheet, and an OTIO cut list.' The README states this greybox clip is the input a named AI video vendor documents wanting: it quotes that vendor's documentation, that a white-model reference video should be used 'as the sole guide for camera movement, pacing, framing, subject motion, and blocking.' That is the video-model vendor's own claim, reported at second hand, not independently confirmed by CozyClay. Its framing - 'export the move, not a prompt' - inverts the ordinary previs relationship: blocked frames become the signal a model is told to follow.
What it is allowed to decide
A CozyClay greybox export can support a class 1 alignment decision - confirming blocking, camera timing and cast positions before generation - because the scene lives in a local, inspectable '.cclayproject' file. That gives it a workable Auditability authority: the blocking behind an export can be reopened and checked. It does not extend that authority to the video model's output: nothing states a generated clip preserves the blocked scene's geometry, camera or timing exactly, so the video should not carry Dimensional or Optical authority merely because a greybox export conditioned it.
The disclosure label
A disclosure label for a shot produced this way, dated 16 September 2026, would read: this is AI-generated video conditioned on a CozyClay greybox export, decision class 1 (alignment), holding an Auditability authority over the blocking stage - the source project file can be reopened - but not Optical or Dimensional authority over the output, since no document states the model preserves the blocked camera or scale exactly. A previs supervisor assembling the pitch is the plausible party asserting this label, since only they hold both the source project and the clip.
- Is the disclosure attached to the blocking stage, the generated clip, or both?
- Can the original .cclayproject file that produced this export still be opened and checked?
- Does the video model's own documentation, not just CozyClay's summary, confirm how closely it follows the reference?
The tool does not claim more than this, and that restraint is notable in a category prone to overstating what a generated clip proves. A locally run export pipeline gives a team a reproducible record of what it asked a model to do; it does not make the answer verifiable.
Sources & reading trail
Repository README states the AGPL-3.0-or-later licence, npx install workflow, and the Export menu's keyframe pack (camera JSON, greybox clip, depth/normal passes), quoting a video-model vendor's own documentation on reading a white-model reference video.
Source published: Not established · Retrieved: 16 September 2026
Project's own site states the block-scene-then-hand-to-AI-video workflow and the phrase 'export the move, not a prompt.'
Source published: Not established · Retrieved: 16 September 2026
Documentation, handbooks, rulings and records establish the entry; the authority reading and the disclosure label are Previs Office editorial analysis. This retrospective draft does not imply the site published on the event date.