Verified Agent Skill record

muapi-storyboard-to-cooking-video

This skill documents a two-step workflow for turning one person-photo URL into a short cooking tutorial: first generate a wide composite reference sheet, then submit that sheet together with the original photo to a reference-to-video model. Treat identity fidelity, shot continuity, and directed audio as requested creative targets that require review, not guaranteed outcomes.

Tasks

Create a composite production reference sheet from one person-image URL, including character views, a kitchen reference, and a cooking action board.
Use the original person image and the completed reference-sheet URL as two references for a cooking-video generation request.
Adapt the documented prompts for a selected dish, kitchen style, outfit, duration, aspect ratio, and resolution.

Inputs

`person_image`: required image URL used as a reference in both generation steps.
`dish`: optional text; defaults to `fresh pasta`.
`kitchen_style`: optional text; defaults to `warm rustic-modern Italian`.
`outfit`: optional text; defaults to `white t-shirt, olive green apron, dark trousers`.
`duration_seconds`: optional integer; defaults to `15`; the document also discusses `10` seconds and adaptations for pieces of 8 seconds or less.
`aspect_ratio`: optional text; defaults to `16:9`; `9:16` is identified for vertical output.
`resolution`: optional text; defaults to `720p`; documented options are `480p` and `720p`.

Outputs

A PNG composite reference sheet and its output URL.
A generated cooking video and its final URL.
An audio-generation request as part of the video call; the document does not establish that requested ambience, effects, timing, or synchronization will be reproduced reliably.

Limitations and checks

The prompts request exact identity preservation, but the document provides no guarantee that a generated reference sheet or video will preserve a face with 100% accuracy.
The prompts request a continuous flow without jump cuts, but the document provides no guarantee that the generated video will form a true single continuous shot or avoid visual discontinuities.
A 30-second extended cut is mentioned only as a follow-up offer; supported commands, parameters, or endpoint limits for a 30-second generation are not documented.
The document asserts support for real human faces but does not document the applicable API moderation policies or establish unrestricted face-generation support.
Audio generation is requested with `--generate-audio true`, but deterministic soundtrack content, synchronization, and adherence to the requested ambience and effects are not established.
The document states that smaller reference-sheet resolutions can blur the action panels and that compressing nine beats into eight seconds or less can degrade motion quality.
Only `480p` and `720p` are listed as video-resolution options.
Confirm that Step 1 uses `gpt-image-v2-edit`, the original person image, a `3840x2160` image size, and PNG output.