15-Second Model Shoot (540p) · عوامی Open Skill

15-Second Model Shoot (540p)

15-Second Model Shoot (540p) Digen پر ایک عوامی Open Skill ہے۔ کھولیں اور دیکھیں یہ ورک فلو کیا تخلیق کرتا ہے۔

@wwws

Custom

19 1 8

https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4284863/2026-10-19/69cre9c7.jpg

---

allowed-tools:
  - recognize_image
  - generate_image
  - generate_video
  - mix_clips
---

# 15-Second Model Shoot (540p)

Turn one uploaded reference photo into a roughly 15-second, portrait, studio "model shoot" video: the same person, the same outfit and the same lighting, shot as a short editorial sequence of slow camera moves. This skill has no structured project state — everything it needs lives in the conversation, in the generated canvas assets, and in the reference files below.

## When to use

- The user uploads a photo and asks for a short video of it (model shoot, campaign clip, pose video, "make this move", "15 seconds").
- The user wants a revision of an already generated shoot clip: a different camera move, slower pacing, another final pose, more focus on the outfit or accessories.
- The user wants a second take from the same reference image.

Do not use it for: single still-image requests only, multi-scene storytelling, dialogue scenes, or edits of footage the user shot themselves (that is a general video-editing task, not this skill).

## Inputs and outputs

- Input: at least one reference image containing one main person. If the user supplies no image, ask for one and stop.
- Output: one portrait 9:16 short video of about 15 seconds, assembled from short reference-conditioned clips, plus (internally) a normalized studio key visual and the individual shot clips.
- Default parameters: `orientation='portrait'`, `resolution='720P'` (the platform tier closest to the app's 540p target), total length about 15 seconds, no audio track unless the user asks for one.

## Reference files

| File | What it holds | When to read |
|---|---|---|
| `references/style-and-prompt.md` | Style, subject/content, layout, must-keep and must-avoid locks, copy-ready English prompt templates | Before any image or video generation |
| `references/video-script.md` | Shot-script structure and how to fill it from this specific request | Before writing the video prompts |
| `references/video-content-constraints.md` | Shot grammar, camera and motion limits, duration feel, forbidden content | Alongside the script, before generating clips |
| `references/workflow-stages.md` | Stage-by-stage sub-steps, checks, retry branches, assembly rules | At the start of the task and whenever a check fails |

Read only the file the current stage needs — do not load all of them at once.

## Steps

1. Start `write_todos` with the stages: reference analysis → key visual → script → clips → assembly → delivery.
2. Call `recognize_image` on the uploaded reference with a prompt asking for: identity cues (face shape, eye shape, makeup style), hair color/length/styling, every garment piece, every accessory and its exact placement, dominant palette, framing and background. Record the result in your reply text so it survives later turns.
3. `read_skill_file` on `references/style-and-prompt.md` and follow its locks. If the photo is low-resolution, cluttered in the background, or badly lit, normalize it first: `generate_image(prompt=<key-visual template filled with the recognized details>, input_urls=[reference], model='i2i.hd.lite', orientation='portrait', name='<subject> — studio key visual', entity_type='characters')`. Use this output as the reference image for every clip. If you skip normalization, use the original upload directly.
4. `read_skill_file` on `references/video-script.md`, then write the full shot script for **this** request: logline, 15-second timeline, 3 beats with shot size, camera move, subject action, expression, hair/wardrobe behaviour, light and transition. Never reuse the worked example verbatim; derive it from the user's wording and the reference analysis.
5. `read_skill_file` on `references/video-content-constraints.md` and check the script against it (shot sizes, motion speed, continuity, forbidden content). Fix the script if anything violates it.
6. For each beat, call `generate_video(prompt=<script beat compressed into one prompt>, input