Anime Character Story Studio · skill pública Open Skill
Anime Character Story Studio
Anime Character Story Studio é uma Open Skill pública no Digen. Abra para ver o que este fluxo cria.
Anime
4 0 3
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4284865/2026-10-21/xb7qwjkg.jpg
# Anime Character Story Studio
Turn one uploaded image into an anime character story: read who the character is, then
produce a small set of original story panels (images) in a consistent anime style that keeps
the character's face, hair and outfit intact. Output is images only — there is no structured
project state; the story plan lives in your own Markdown reply plus the generated panels.
## When to use
- The user uploads an image and asks for anime / manga versions, a "character story",
story panels, scenes, or an episode of that character.
- The user asks to identify the anime character in an uploaded picture and build a story
around them.
- The user wants only one more segment, one re-drawn scene, or a correction to a single
panel of a story you already produced.
If the user gives no image at all, ask them to upload one before generating — this skill is
reference-driven and must not invent a character from scratch.
## Inputs and outputs
- Input: one reference image (uploaded asset, canvas asset, `asset_id`, `s3://` uri, or a
preset-asset key). A second image may be supplied as supporting reference (outfit, prop,
background); treat it as secondary and never let it override the main character's identity.
- Output: a set of story panels as images (`entity_type='story_segments'`), portrait by default.
## Steps
1. **Read the reference image.** Look at it first: hairstyle and hair colour, eye colour,
face shape, outfit and its material/accents, signature props, signature palette, pose and
expression, and the setting it implies. Only call `recognize_image(image_url=...)` when the
image is unclear, cropped, or text-heavy and you need a second opinion — never use it as a
substitute for actually looking at the picture. Compress the result into a short "identity
anchor" line (hair, eyes, outfit, palette, temperament) in your reply; this exact line is
reused in every segment prompt.
2. **Read the visual conventions.** Call
`read_skill_file(name='anime-character-story-studio', path='references/style-and-prompt.md')`
and follow it. It holds the style rules, subject/content rules, layout rules, must-keep and
must-avoid lists, and the copy-ready English prompt template. Do not write generation
prompts before reading it.
3. **Read the story rules.** Call
`read_skill_file(name='anime-character-story-studio', path='references/story-segments.md')`
and follow it. It defines how many segments to make, how to derive personality from visual
cues, how the arc should escalate, and how each segment prompt is filled in.
4. **Plan, and show the plan.** For a multi-segment request, call `write_todos` with one todo
per segment (content = segment title). Then state the plan briefly in user-facing language:
character identity anchor, segment count, one line per segment beat. Do not stop for
confirmation unless the user asked to review first.
5. **Generate each segment — one `generate_image` call per panel.** Before the first call,
build the prompt from the template in `references/style-and-prompt.md`, filling the identity
anchor, the segment beat, and the layout rules from `references/story-segments.md`.
Call:
- `generate_image(prompt=<English prompt built from the template>, input_urls=[<reference image>], model='i2i.hd.lite', orientation='portrait', name=<short segment title>, entity_type='story_segments')`
- `model='i2i.hd.lite'` is the default and must be passed for every panel — this is a
reference-driven redraw, not text-to-image.
- Always pass the **same** reference image in `input_urls` for every segment so the
character stays identical across panels. If the user supplied a supporting reference,
append it as a second entry.
- `orientation='portrait'` unless the user explicitly asks for landscape or square; leave
`resolution` unset (default 720P) unless the user asks for a specific quality tier.
- Keep each `name` short and distinct (p