Bathroom Selfie Video · skill pública Open Skill
Bathroom Selfie Video
Bathroom Selfie Video é uma Open Skill pública no Digen. Abra para ver o que este fluxo cria.
Other
0 1 3
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/3656543/2026-10-14/ktdbzcud.jpg
# Bathroom Selfie Video Fixed pipeline: **lock the user’s character** → **text-to-image bathroom environment** (`t2i`) → **one 5-second reference-to-video clip** (`ref2v`) from a **selfie-device camera POV**. No storyboard grid, no multi-shot edit, no `mix_clips` unless the user explicitly asks to stitch unrelated clips. ## When to use - User uploads a person photo and wants a **bathroom selfie-style** vertical video. - User wants **only this workflow** (not a general movie maker or multi-scene short). - User asks to **regenerate the bathroom still**, **re-lock character**, or **re-render the 5s video** without changing the overall pattern. ## Steps（步骤） 1. **Lock character** — `read_skill_file` → `references/01_lock_character.md`; store `character_ref_url` + identity tags (`recognize_image`). 2. **Bathroom scene** — `references/02_bathroom_scene_t2i.md`; `generate_image` with **`model='t2i'`**, no people in frame. 3. **5s selfie clip** — `references/03_ref2v_selfie_video.md`; `generate_video` with **`model='ref2v'`**, **`duration=5`**, single continuous selfie POV with slow vertical camera tilt. Orchestration and Project brief: `references/00_orchestrator.md`. ## Production stages（阶段 / references/） Read **one** file per stage via `read_skill_file(bathroom-selfie-video, path)`. Do not load all references at once. | Stage | File | Tools | |-------|------|-------| | Orchestration | `references/00_orchestrator.md` | write_todos, set_guidance | | 1 Lock character | `references/01_lock_character.md` | recognize_image | | 2 Bathroom scene | `references/02_bathroom_scene_t2i.md` | generate_image (`t2i`) | | 3 Selfie video | `references/03_ref2v_selfie_video.md` | generate_video (`ref2v`) | ## Collaboration - **Default flow:** portrait **9:16**, **5s**, **single continuous shot**, **selfie / front-camera POV** (device held at arm’s length or mirror-phone setup—see stage 3). **No shot cuts**, no angle jumps; only **slow tilt/pan up or down** within the same take. - **First turn:** If the user already attached a person image, go to stage 1 immediately. If not, ask for **one clear photo of the subject** (face visible, good lighting) before any generation. - **One-click:** After character lock, run stage 2 then stage 3 **without** asking between stages unless the user said “wait” / “confirm first”. - **Partial rerun:** User wants new bathroom only → stage 2 + 3. User wants new motion/prompt only → stage 3 (reuse locked character URL + latest bathroom still URL). - Start multi-step work with `write_todos`; call `set_guidance` when the 5s clip is delivered or when blocked (missing upload, policy). - On tool failure: report honestly; never fabricate media URLs. ## Canvas - **Character lock:** user upload stays primary; optional `characters` asset only if you must duplicate for canvas clarity—prefer citing the user attachment. - **Bathroom still:** one `locations` image (`name`: short label e.g. `Bathroom scene`). - **Final:** one `final_videos` asset (`name`: e.g. `Bathroom selfie – 5s`), **5 seconds**, portrait. ## Content policy - **SFW only:** everyday bathroom selfie (skincare, outfit check, casual vlog energy). **Reject** nudity, explicit sexual content, minors in sensitive contexts, or voyeuristic angles. Do not use generation tools for disallowed requests. ## Language - All **image and video prompts** in **English**. - User-facing chat may follow the user’s language; keep prompt fields English.