Who want to fight · öffentlicher Open Skill

Who want to fight

Who want to fight ist ein öffentlicher Open Skill auf Digen. Öffne ihn, um zu sehen, was dieser Workflow erstellt.

@JackRhino

Other

4 1 6

https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4284865/2026-10-20/vazj3gak.jpg

# Who want to fight

From the user’s text and/or reference images, produce **time-segmented fight prompts** (single-frame decisive moment + video beats), then a **contact-moment storyboard still**, then a **ref2v** video with grounded weight and practical stunt feel. Any number of fighters; choreography follows the user’s request.

## When to use

- User describes two-or-more close-quarters combat, punches, kicks, throws, or weapon exchanges, or uploads character/scene references for a fight shot.
- User asks for fight prompts, martial-arts storyboards, fight video generation, or edits to one move, one shot, duration, or orientation.
- User wants realistic contact, practical choreography—not wire-fu, not cartoon.

## Production stages (references/)

**Do not read every reference at once.** Pick the current stage and use `read_skill_file` for that file only.

| Stage | File | Tools | When to read |
|------|------|------|--------|
| Orchestration | `references/00_orchestrator.md` | write_todos, set_guidance | Multi-step start, stage gates |
| Intake & plan | `references/01_intake_plan.md` | recognize_image | First turn, clarify count/style/duration |
| Prompt authoring | `references/02_prompt_authoring.md` | create_document / update_document | Before writing detailed fight prompts (required) |
| Keyframe still | `references/03_keyframe_image.md` | generate_image | Before contact-moment image |
| Fight video | `references/04_fight_video.md` | generate_video, mix_clips | After still is ready |

## Collaboration rules

- **One-click pipeline**: Follow the table—plan → read `02` and write prompts (optional `create_document` titled fight prompt pack for the user) → keyframe → video. **Do not pause for confirmation** between stages unless the user says “show me the prompt first” or “wait for my OK”.
- **Single-stage edit**: Read only the matching reference; do not rerun the full pipeline.
- **Fighter count & segments**: No cap on fighters; long combos or multiple matchups → multiple ref2v segments; stitch with `mix_clips` when a continuous reel is needed (see `04`).
- Call `write_todos` at multi-step start; call `set_guidance` after plan, prompt delivery, keyframe, or final output.

## Generation channels (hard-coded; no physical model names)

| Use | Channels (choose only from this set) |
|------|--------------------------------------|
| Text-to-image | `t2i` / `t2i.hd` / `t2i.hd.lite` / `t2i.hd.gpt` |
| Image-to-image | `i2i` / `i2i.hd` / `i2i.hd.lite` / `i2i.hd.gpt` |
| Fight video | **`ref2v`** (at least one reference still in `input_urls`) |

Defaults: no reference image and quality matters → `t2i.hd`; user or prior character still → `i2i.hd`; fast draft → `.lite`; user asks for GPT channel → `.gpt`. `.hd*` does not support NSFW.

## Canvas naming (entity_type)

- Standoff / keyframe stills: `entity_type='storyboard_images'`, `name` e.g. `Standoff-pre-contact`, `A-cross-B-block`.
- Fight clip: `entity_type='storyboard_videos'`, `name` e.g. `Seg1-cross-block`, `Final-clash`.
- Multi-segment stitch: `mix_clips` → `entity_type='final_videos'`, `name` `{user title} - fight reel`.
- Prompt document: `create_document` → `entity_type='scripts'`, `name` `Fight prompts - {title}`.

## Content boundaries

- Tone: **PG-13 hard-realistic martial arts**—dust, sweat, abrasions, bruises OK; **refuse** dismemberment, gore, excessive blood, torture, violence targeting real identifiable people.
- Video generation does not support NSFW/adult content.

## Notes

- If a tool returns `success=false`, report honestly; never fabricate URLs.
- Video prompts must use **0.3–0.5s beats** and full body-chain detail (see `02`); never summarize a full combo in one sentence.
- Final replies summarize in natural language; put long prompts in a document or a short summary—do not dump walls of text in chat.