One-Take Pullback Blue Planet Reveal · publiczny skill Open Skill

One-Take Pullback Blue Planet Reveal

One-Take Pullback Blue Planet Reveal to publiczny Open Skill w Digen. Otwórz go, by zobaczyć, co tworzy ten workflow.

@Promise

News

0 0 2

https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/3949831/2026-10-19/4qxsm4j2.jpg

# One-Take Pullback Blue Planet Reveal

Turn one uploaded portrait photo into a vertical short video: a single continuous 8-second pull-back that starts as a close-up of the person, keeps pulling back past the city and the atmosphere, and ends on the revealed blue planet. The identity of the uploaded person must never change; everything else (wardrobe, pose, scene, palette, light) is re-styled into the single locked "pull-back blue planet" look.

## When to use

- The user uploads a portrait/photo and wants this pull-back reveal short video.
- The user wants to change one part of an already generated pull-back video (keyframe look, wardrobe/scene, motion speed, duration, orientation).
- Not for: generic single-image generation, unrelated video editing, multi-shot story videos, or anything that needs several cuts.

## Inputs and output

- **Reference image (required)** — the person whose identity is locked. It is the `input_urls` of the image-to-image keyframe step and the first frame of the video step.
- **Text (optional)** — extra wishes such as a scene hint, wardrobe direction, mood or pacing note. Fold them into the prompt, but never let them override the identity lock, the locked blue-planet look, or the one-take rule.
- **Output** — one short video: portrait 9:16, 1080P, about 8 seconds, one continuous pull-back.

## Reference files

| Stage | File | Tools | Read when |
|---|---|---|---|
| Intake, step details, checks, retries | `references/workflow-stages.md` | — | at the start of every request |
| Look, subject rules, prompt fragments | `references/style-and-prompt.md` | `generate_image` | before writing the keyframe prompt |
| Shot, motion, prohibitions, delivery | `references/video-content-constraints.md` | `generate_video` | before writing the video prompt |

Read one file at a time with `read_skill_file`; never load all files up front. If only one part of the video is being revised, read only the file for that part.

## Steps

1. Call `read_skill_file` for `references/workflow-stages.md` and follow its intake checks: confirm the reference image exists, extract any text hints, decide orientation and duration, and set up a `write_todos` list (intake → keyframe → pull-back video → delivery) when more than one stage will run.
2. **Stage 1 — styled keyframe.** Call `read_skill_file` for `references/style-and-prompt.md`, build the image prompt from its template and fragments, then call:
   `generate_image(prompt=<built image prompt>, input_urls=[<the uploaded reference image>], model='i2i.hd.lite', orientation='portrait', resolution='1080P', name='Blue Planet Keyframe', entity_type='keyframes')`.
   Run the keyframe checks in `references/workflow-stages.md` before continuing (identity intact, headroom kept, unified palette, no text/watermark).
3. **Stage 2 — one-take pull-back video.** Call `read_skill_file` for `references/video-content-constraints.md`, build the motion prompt from it, then call:
   `generate_video(prompt=<built motion prompt>, input_urls=[<the keyframe asset id or URL returned in step 2>], model='i2v', orientation='portrait', resolution='1080P', duration=8, name='Blue Planet Pullback', entity_type='final_videos')`.
   Never invent a URL: use only the asset id/URL returned by the previous tool call.
4. Verify the result against the delivery checks (single continuous shot, no cut, identity stable, blue planet revealed at the end). Follow the retry rules in `references/workflow-stages.md` if a check fails.
5. Reply in natural language: what was produced, which direction/duration was used, and what can be adjusted. Do not paste long URLs. If you offer next steps, call `set_guidance` in the same turn and call no other tool in that turn.

## Notes

- Every generate step uses a stable model channel — `i2i.hd.lite` for the styled keyframe, `i2v` for the pull-back video. Never write a physical model name.
- Default framing is portrait 9:16 at 1080P and about 8 seconds; only switch to landscape or another duratio