Anime POV Scene Maker · skill Open Skill công khai
Anime POV Scene Maker
Anime POV Scene Maker là một Open Skill công khai trên Digen. Mở ra để xem workflow này tạo gì.
Anime
8 3 5
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4166649/2026-09-23/3emg6z8b.png
# Anime POV Scene Maker
Turn a one-line idea into a short, immersive **first-person (POV) anime video scene** — the viewer feels like they've stepped into an anime world. Built for creators to reuse in Shorts, Reels, edits, and storytelling videos.
## When to use
- The user wants a short anime POV video scene (setting + character + situation + action).
- The user wants to change one part of an existing scene (new action, restyled character, different setting, redo the clip).
## Reference files
- `references/pov-prompt-guide.md`: How to construct an immersive POV prompt (camera, sensory detail, beginning/movement/ending). Read it before writing any prompt.
## Memory (cross-session preferences)
This skill remembers the user's preferred anime style and recurring characters so future scenes match them automatically.
- **When to query**: at the start of a new scene, call `memory_query` (e.g. prompt "user's preferred anime style, favorite settings, recurring characters") before filling in defaults. If nothing relevant is found, don't invent preferences — just use sensible defaults.
- **What to store**: durable preferences worth reusing across sessions — preferred art style (e.g. "90s cel-shaded", "modern high-detail", "pastel soft"), favorite settings, and any explicit "always/never" rules the user states. Use `memory_store` with `kind='preference'` (or `'project'` for a recurring series). Use a `namespace` matching the skill name (`anime-pov-scenes`) to keep writes isolated. Don't store one-off scene details or temporary drafts.
- **Recurring characters**: when the user asks to save a character (e.g. "save this character", "make her a recurring character", "remember this swordsman"), store a full character profile with `memory_store` using `kind='person'` and a `namespace` of `anime-pov-scenes`. Include: name (if any), appearance (hair, eyes, outfit, distinguishing features), personality, and any style notes. When a scene references a saved character, `memory_query` for it and reuse that profile in the prompt so the character stays consistent across scenes.
- **When to store**: after a scene is made, if the user expresses a preference ("I love this style", "make it always like this") or asks to save a character, store it. Also store any style/character the user explicitly asks to remember.
- **Forget/correct**: when the user says "forget/change my style" or "delete this character", call `memory_query` to find the item's `id`, then `memory_forget` on it.
## Steps
1. **Clarify the scene (4 parameters).** First call `memory_query` to recall the user's preferred anime style, favorite settings, and any saved recurring characters. Then ask for anything not yet given:
- **Setting** — where the viewer is (e.g. neon Tokyo alley at night, cherry-blossom shrine, fantasy academy rooftop, cyberpunk train).
- **Character** — who is in front of the viewer (e.g. a silver-haired swordsman, a shrine maiden, a cat-eared classmate). If the user names a saved recurring character, use that saved profile. If none given, default to a striking anime character that fits the setting.
- **Situation** — the moment (e.g. they just saved you, a confession, a battle standoff, a festival).
- **Action** — what happens in the clip (e.g. they reach out a hand, draw a blade, turn to face you, offer a flower).
If the user gives only a one-liner, fill in the rest with a strong default (preferring any remembered style/character) and tell them what you chose (they can change it).
2. **Write the POV prompt.** Read `references/pov-prompt-guide.md`, then write a detailed English prompt: first-person camera, the setting, the character, the situation, and a clear beginning → movement → ending. Fold in the user's remembered style preferences and any saved character profile. Keep it SFW (video generation does not support NSFW/adult content).
3. **Generate the POV keyframe image** with `generate_image`:
- `model='t2i'`, `orientation='portrait'` (9:16, fo