Anime Character Story Workshop · Open Skill 公開スキル

Anime Character Story Workshop

Anime Character Story Workshop は Digen 上の公開 Open Skill です。開くと、このワークフローが何を作るか分かります。

@JackRhino

Anime

2 0 3

https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4284865/2026-10-21/hfdrry9q.jpg

# Anime Character Story Workshop

Turns one uploaded anime character image into three deliverables that all describe the same character:

1. a short character profile,
2. an original short story written around that character,
3. an original anime illustration (character story card / key visual) that keeps the uploaded character on-model.

There is no structured project state. The profile and the story live in your reply text; the artwork becomes a canvas asset.

## When to use

- The user uploads an anime character image and asks for a character setting, profile, backstory, or story.
- The user uploads an anime character image and asks for an illustration, card, poster, or key visual of that character.
- The user wants to revise one part only (reword the profile, continue the story, regenerate the image) for a character already uploaded in this conversation.

## Steps

1. **Read the uploaded reference image.** Look at the image in the conversation directly. If the character details are still ambiguous, or the image is only available as an asset reference, call `recognize_image` on it and ask for a structured description of hairstyle, eye color and shape, facial features, outfit, accessories, and background palette. Run the reference inspection checklist in `references/step-details.md`.

2. **Load the writing and prompt rules.** Call `read_skill_file` for `references/style-and-prompt.md` and `references/character-story-spec.md`. Do not build the prompt or the story before reading them.

3. **Build the must-keep list.** Write down, in the reply or in your working notes, the identity anchors that may never change: hair length/color/bangs, eye color, face shape, outfit design and its color blocking, accessories, and the anime medium itself.

4. **Write the character profile and the original story** following the structure, length, and tone rules in `references/character-story-spec.md`. Derive the setting from visible cues in the reference background, and reuse at least two visible design details inside the story so text and artwork stay consistent. Keep the story original — never reproduce an existing franchise or a known character's plot.

5. **Compose the image prompt** from the templates in `references/style-and-prompt.md`: fill in the character fields, the pose, the scene, the holographic label text, and the identity-consistency clause. Use the negative prompt from the same file. Keep the prompt in English.

6. **Generate the illustration** with the reference image as the image input:

   `generate_image(prompt=<prompt from step 5>, input_urls=[<uploaded reference image asset_id / s3:// uri / https url>], model='i2i.hd.lite', orientation='portrait', name='<short character name> - story card', entity_type='characters')`

   - `model='i2i.hd.lite'` is the fixed channel for this app: it is a reference-image-driven image. Never pass a physical model name.
   - `input_urls` must contain the uploaded reference image. Never invent an image reference.
   - `orientation='portrait'` by default; switch to `landscape` or `square` only when the user explicitly asks.
   - Do not pass `resolution` unless the user asks for a specific sharpness; the platform default is fine.
   - If the user asks for a scene from the story instead of a card, use the scene template in the same reference file and keep the same channel and inputs.

7. **Verify the result** against the must-keep list using the post-generation checks in `references/step-details.md`. If the face, hairstyle, outfit, or palette drifted, regenerate once with a stronger identity clause, then stop and report honestly.

8. **Reply with the profile and the story in natural language**, then mention that the illustration was generated and is on the canvas. Do not paste long URLs or raw JSON. If the user only asked for text, skip steps 5-8's image step entirely and deliver the profile and the story only.

9. Optionally call `set_guidance` in the same turn as the final reply, with next-step op