Halloween Cosplay Reference Image & Short Video · Open Skill 公开技能
Halloween Cosplay Reference Image & Short Video
Halloween Cosplay Reference Image & Short Video
TryOn
0 0 0
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4150474/2026-10-15/m3hawrrf.jpg
# Halloween Cosplay Reference Image + 7-Second Short Video Turn the user's uploaded portrait photo into a Halloween cosplay look, producing **1 cosplay reference image** (optionally a three-view character sheet) and **1 ~7-second cosplay short video**. The person's identity must be preserved — only the outfit / makeup / accessories / scene change. There is no structured project state; progress lives in the conversation + canvas assets + `write_todos`. ## When to use - The user uploads a portrait photo and asks for a Halloween cosplay look. - The user wants a cosplay reference image (three-view / character sheet) or a few-second on-camera cosplay clip. - The user gives one of the Halloween performance concepts numbered 0–9 (see `references/concepts.md`). - Also applies when the user wants only one of the two outputs (reference image only / video only). ## Inputs and prerequisites - **Required**: one portrait photo (an image asset uploaded by the user this turn, or an existing image asset on the canvas). `asset_id` / `s3://` / https can all be passed directly as `input_urls` — do not convert formats yourself. If the user uploads several, pick the one with a clear frontal face and even lighting as the **single** person reference; if unsure, ask one short question. - With no portrait photo, **ask the user to upload one first**; never invent a person from nothing. - **Optional**: performance concept number 0–9. If the user hasn't specified one, use `set_guidance` to offer 3–4 visually distinct recommendations (for example 0 / 3 / 6 / 8) and let the user reply with a number — do **not** turn all 10 concepts into buttons. - The uploaded photo does not need to be a studio-quality reference (a phone snapshot or a screenshot is fine). ## Steps 1. **Confirm the person reference**: lock onto the uploaded image asset; both the reference image and the video use it as the only source of the person. 2. **Determine the performance concept**: once the user gives a number, call `read_skill_file(name='halloween-cosplay-video', path='references/concepts.md')` and take only the fields (`OUTFIT` / `SCENE` / `LIGHTING-PALETTE` / `MOTION_7S`) of **that one card** — do not import details from other concepts. If the user has not chosen one → use `set_guidance` as described in the previous section. 3. **write_todos**: list "cosplay reference image → 7-second short video" (keep only one item when the user wants just one), refreshing the whole table as status changes. 4. **Generate the cosplay reference image** (default first step): - `generate_image`, `input_urls=[portrait photo]` - `model='i2i'` (there is a reference image; use the **bare `i2i`**, not a `.hd*` variant) - `orientation='portrait'` (matching the short video); switch to `orientation='landscape'` only when the user explicitly wants a **three-view side-by-side sheet** - `name='<short concept name> reference'` (e.g. `Witch Village reference`), `entity_type='characters'` - prompt: apply the "Reference image prompt skeleton" below, filling in the card's `OUTFIT` / `SCENE` / `LIGHTING-PALETTE` - For a three-view request, append to the skeleton: `three-view character sheet: front view, side view and back view of the same outfit, arranged side by side in one frame, consistent face and hairstyle across all three views` 5. **Generate the 7-second short video** (right after the reference image succeeds — do not keep asking for confirmation): - `generate_video`, `input_urls=[the reference image from the previous step]` (the first image is the first frame / reference) - `model='i2v'` (image-to-video; video generation does not support NSFW/adult content) - `orientation='portrait'`, `duration=7`; if the platform clamps duration/resolution, tell the user the actual value from the returned `param_adjustments` instead of insisting on 7 seconds - `name='<short concept name> 7s'`, `entity_type='final_videos'` - prompt: apply the "Video prompt skeleton" below, filling the