Car Reel editor · Open Skill 공개 스킬

Car Reel editor

Edit your car into viral shots

@aadityajaat

Other

7 5 8

https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/1298230/2026-10-16/eg3q2cnq.jpg

---

allowed-tools:
  - recognize_image
  - generate_video
  - mix_clips
  - generate_music
  - generate_tts
  - generate_image
  - list_models
  - search_library
  - get_library_item
  - split_image
  - recognize_video
  - transcribe_audio
  - fetch_video_info
  - download_video
  - create_document
  - update_document
  - get_document
  - memory_store
---

# Car Reel Director — Single Image → Cinematic Car Reel

## When to use
- The user uploads ONE photo of their car and wants a premium cinematic automotive reel (vertical 9:16, Instagram/Reels aesthetic).
- The user wants to redo/refine one shot, or re-assemble the full reel from the same master image.

## Reference files
- `references/shot-director.md`: the full playbook — 10-shot camera language, lighting system, VFX, transitions, continuity checklist, and the negative prompt. **Read it before generating any shots** and follow it for every `generate_video` prompt.

## Core principle — identity lock
The uploaded image is the **MASTER REFERENCE**. The exact car in it must remain identical in every shot: same model, body shape, proportions, paint color, wheels, headlights/grille/bumpers/mirrors/windows/spoiler/badges/license plate. Never redesign, replace, or morph the car. Camera, lighting, environment and VFX may change — the car itself must not.

## Steps
1. **Analyze the master image.** Call `recognize_image(image_url=<master asset>, prompt="Describe the vehicle in detail: exact model, body style, paint color, wheel design, headlight/grille/bumper/mirror/window/spoiler/badge details, license plate, and the camera angle/orientation (front/rear/side/three-quarter).")`. Use this to lock the identity and to write accurate shot prompts.
2. **Read the playbook.** `read_skill_file(name="car-reel-director", path="references/shot-director.md")` and use it for every shot prompt.
3. **Generate the 10 shots.** For each shot, call `generate_video` with:
   - `model='ref2v'` (reference-to-video — keeps the car consistent from the master image)
   - `input_urls=[<master image asset_id>]` (the SAME master image for every shot)
   - `orientation='portrait'` (9:16 vertical reel)
   - `resolution='1080P'`
   - `prompt` = the shot's camera-language prompt from the playbook, plus the identity-lock line and the negative prompt.
   - `name` = the shot title (e.g. "Shot 01 — Establishing Reveal"), `entity_type='car_reel_shots'`.
   Generate shots one at a time (or in small batches), checking each result against the master before moving on; if a shot drifts from the car, regenerate it with a stronger identity-lock prompt.
4. **Optional soundtrack.** If the user wants music, call `generate_music(prompt=<cinematic automotive score description>, name="Car Reel Score", entity_type='music')`.
5. **Assemble the reel.** Call `mix_clips` with one clip per shot in order (`clips[].video.url=<shot asset_id>`), `orientation='portrait'`, `name="Cinematic Car Reel"`, `entity_type='final_videos'`. If a soundtrack was generated, add it as an audio track (`clips[].audio.url=<music asset_id>`).
6. **Reply.** Summarize in natural language: the shots generated, the final reel, and any shot that had to be regenerated. Don't paste raw URLs.

## Notes
- The master image is the identity anchor — always pass the same asset_id to every `generate_video` call.
- `ref2v` requires at least one reference image; the master image satisfies this. Do not pass the master image as a start/end frame — it is a reference.
- If a shot comes out with a wrong car, deformed body, or changed wheels/paint, regenerate it with a stricter identity-lock prompt rather than accepting it.
- Report failures honestly; never fabricate URLs. If the input isn't a clear single-car photo, ask the user to confirm before generating.
- Intermediate shot clips will appear on the canvas alongside the final reel — this is expected.