Dynamic3D Reels · skill pública de Open Skill

Dynamic3D Reels

Dynamic3D Reels es una Open Skill pública en Digen. Ábrela para ver qué crea este flujo.

@ronialt

Meme

4 0 2

https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/3487427/2026-09-30/94pm5wtg.png

# Dynamic3D Reels

Turn a one-line idea, a draft script, or a reference image into a fast, punchy, very dynamic **3D-style vertical Reel** (9:16 portrait). The look is a consistent, vibrant 3D-render aesthetic (stylized, high-energy, cinematic lighting) with quick cuts and rhythmic pacing. Keep every stage fast and snappy — the user wants speed and dynamism, not long slow scenes.

## When to use
- User wants a 3D dynamic vertical Reel/short from an idea, script, or reference image.
- User wants to redo or tweak one stage: rewrite the script, generate the next clip, restyle a character, or mix the episode.

## Reference files
- None required. (This skill is self-contained.)

## Steps

1. **Clarify the brief (briefly).** Ask only what's missing: the core idea/subject, desired length (default ~15–30s), and any character/style reference. If the user gave a one-liner, proceed — don't over-interrogate. If they uploaded a reference image, use it to lock the character/subject style.

2. **Write the script.** Draft a short, fast-paced vertical script (hook in the first 2 seconds, escalating beats, punchy ending). Save it with `create_document(content=<script>, name="Reel Script", entity_type="scripts")`. Show the user the script and confirm before generating visuals. If they only want a rewrite, use `update_document` on the existing script asset instead.

3. **Generate storyboard images (3D style).** For each beat, call `generate_image` with:
   - `model='t2i'` (no reference) or `model='i2i'` (with a reference image in `input_urls`).
   - `orientation='portrait'` (9:16).
   - Prompt in English, always emphasizing the 3D style: "stylized 3D render, Pixar-like, vibrant colors, dynamic composition, dramatic cinematic lighting, high energy".
   - `name` = shot title (e.g. "Shot 1 - Hook"), `entity_type="storyboard_images"`.
   - Keep the character/subject consistent across shots by reusing the same reference image or restating the same character description.

4. **Generate video clips.** For each storyboard image, call `generate_video` with:
   - `model='i2v'`, `input_urls=[<storyboard image asset_id>]`.
   - `orientation='portrait'`, short `duration` (2–4s per clip) for fast, dynamic motion.
   - Prompt describing the dynamic motion (camera push-in, whip pan, character action, energy burst).
   - `name` = clip title, `entity_type="storyboard_videos"`.

5. **Optional audio.** If the user wants voiceover, call `generate_tts` with `model='tts.design'`, `voice_instruction` in English describing the energetic voice, `name`/`entity_type="voiceovers"`. If they want background music, call `generate_music` with an energetic 3D-adventure-style prompt, `name`/`entity_type="music"`.

6. **Mix the final Reel.** Call `mix_clips` with the generated clips in order (each as a `video` track), adding the voiceover/music as `audio` tracks as available. Pass `orientation='portrait'`, `name="Final Reel"`, `entity_type="final_videos"`.

7. **Reply.** Summarize the final Reel in natural language (don't paste long URLs). Offer next steps: restyle a character, generate the next clip, rewrite the script, or remix.

## Notes
- Always 9:16 portrait unless the user explicitly asks otherwise.
- Keep the 3D-render style consistent across all shots; restate the same style keywords and character description in every prompt.
- Intermediate products (storyboard images, clips, script) will be visible on the canvas — that's expected.
- Report failures honestly; never fabricate URLs. Politely decline prohibited content (video generation does not support NSFW/adult content on any channel).
- For multi-step runs, first use `write_todos` to list the steps and update it as you progress; when a stage completes, call `set_guidance` to suggest next actions.