Fruit Story Studio · compétence Open Skill
Fruit Story Studio
Fruit Story Studio est une Open Skill publique sur Digen. Ouvrez-la pour voir ce que ce workflow crée.
ShortDrama
1 0 1
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4307538/2026-10-22/hv4t5a3n.jpg
# Fruit Story Studio Turns a one-line idea into a viral-style funny fruit-story animation: anthropomorphic 3D-cartoon fruits competing in an absurd challenge (freezer, hot sauce, talent show, …) with a punchy hook, comedic dialogue, and a "Part 2" cliffhanger — portrait 9:16, about 45–60 seconds, finished with word-by-word highlight captions. The Skill has no structured project state. Progress lives in the conversation, in canvas assets, and in `write_todos`. ## When to use - The user wants a funny animated fruit story video (any fruit, any challenge). - The user mentions "fruit story", "fruit challenge", "talking fruits", "part 2 video", or pastes a reference TikTok/Reels fruit animation. - The user wants to revise one stage of an in-progress fruit story (script, characters, a scene, voiceover, or captions). Do not use for realistic human videos, non-fruit stories, or still-image-only requests. ## Production stages (references/) Detailed steps for each stage live under `references/`. **Do not read them all at once.** Decide the current stage first, then use `read_skill_file` to read only that file. | Stage | File | Tools used | When to read | |---|---|---|---| | 1. Story & script | `references/scriptwriting.md` | `create_document`, `memory_*` | The user gives an idea, or wants the story/dialogue changed | | 2. Character design | `references/character-design.md` | `generate_image` | Character look is needed, or a new fruit joins the cast | | 3. Storyboard stills | `references/visuals.md` | `generate_image` | One still per scene is needed | | 4. Scene animation | `references/animation.md` | `generate_video` | A still must become a moving clip | | 5. Voiceover | `references/voiceover.md` | `generate_tts` | Narration or character dialogue audio is needed | | 6. Music (optional) | `generate_music` directly (`model='music'`) | `generate_music` | User wants background music | | 7. Assembly & captions | `references/assembly.md` | `mix_clips`, `caption_video` | Clips must become one finished video with captions | | Prompt templates | `references/prompt-library.md` | (copy/paste) | You need ready-made prompt recipes for characters, scenes, motion | ## Collaboration principles - If the user says "make the video" or gives a one-line idea: call `write_todos` with the stage list, then advance through the table in order. Sync progress after each stage, then immediately read the next stage file. **Do not ask for confirmation between stages.** Stop only if the user said "confirm first". - If the user only wants one stage changed: read only that file. Do not redo the whole pipeline. - Ask at most one question at the start: the dialogue/caption language (default English). If the user already wrote in a language, use it without asking. - Image and video prompts are always written in English. Dialogue, narration, and captions stay in the user's language. - `orientation='portrait'` everywhere unless the user explicitly asks otherwise. Omit `resolution` unless asked (platform default applies). - Intermediate outputs stay hidden: pass `add_to_canvas=false` for storyboard stills, per-scene clips, and raw voiceover files. Show the user: the script document, the character design sheet, and the final captioned video. - Pass a short `name` on every generation call (character name, scene title, final title) and the stage's `entity_type` convention from its reference file. - Hardcode generation channels exactly as the reference files say (`model='t2i'`, `model='ref2v'`, `model='tts.design'`, `model='music'`, `caption_video` gets no `model`). Never write a physical model name. - When the pipeline finishes a stage boundary or the whole video, call `set_guidance` in the same turn as the final reply with the sensible next actions (e.g. "Make Part 2", "Revise scene 3", "New fruit story"). Never call any other tool in that turn. - `caption_video` can take from the teens of minutes to over twenty. Do not poll while it