Bathtub Bath-Time Photo · публичный навык Open Skill

Bathtub Bath-Time Photo

Bathtub Bath-Time Photo — публичный Open Skill на Digen. Откройте, чтобы увидеть, что создаёт этот процесс.

@wwws

Product

5 0 2

https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4284863/2026-10-20/csqqwqj4.jpg

# Bathtub Bath-Time Photo

An image-to-image skill. It takes one uploaded photo plus optional text instructions and returns a single photorealistic portrait of that same person in a warm bathtub bath-time scene: thick bath towel wrapped around the body, towel wrapped on the head, steam and warm light. Identity, face and structure come from the uploaded photo; clothing, pose and scene are replaced by the bath-time theme.

## When to use

- The user uploads a photo and asks for a "bathtub photo", "bath time photo", "spa / towel-turban portrait", or says "put me in a bathtub".
- The user wants to regenerate, restyle or adjust a bath-time portrait that was already produced in this conversation.
- The user asks for a portrait remake where the scene must be a bathtub and the styling must be towel + steam + warm water.

Do not use this skill for unrelated edits (background swap in a street scene, product shots, group/party scenes, anime illustration, video output).

## Inputs and outputs

| Item | Convention |
|---|---|
| Reference image | Required in the normal flow. One clear photo with a visible face, preferably one person. Use the canvas asset id / `s3://` uri / https url provided in this conversation. |
| Text instruction | Optional. Free-form styling wishes (warm tone, smiling, hair hidden, half-body, spa candles, etc.). Only theme-compatible wishes are applied. |
| Output | One image asset on the canvas. No video, no audio. |

## Steps

1. **Collect inputs.** Take the reference image from the user's upload / canvas asset and read any text instruction. If there is no image at all, ask for one photo (single person, face clearly visible, not blurry). Do not invent an image url.
2. **Inspect the photo when needed.** If the photo's subject count, face visibility or image quality is unclear, call `recognize_image` on the reference and note the identity anchors (gender presentation, approximate age, face shape, skin tone, eye/brow/nose/lip shape, expression, any distinctive marks). Skip this call when the photo is obviously a clean single-person portrait.
3. **Read the visual rules.** Call `read_skill_file` for `references/style-and-prompt.md` and follow its Style / Subject / Layout / Must keep / Must avoid rules and its prompt template. The prompt must be built from that file, not improvised.
4. **Read the step details.** Call `read_skill_file` for `references/step-details.md` and apply its pre-checks, branching rules, parameter table, verification and retry ladder.
5. **Write the English prompt.** Fill the template from `references/style-and-prompt.md`: identity anchors from the uploaded photo (step 2), the bath-time theme block, and only the theme-compatible parts of the user's text instruction. Never translate the prompt into another language.
6. **Generate.** Call `generate_image` with:
   - `prompt` = the assembled English prompt;
   - `input_urls` = `[reference image]` (asset id / `s3://` uri / https url from this conversation);
   - `model` = `i2i.hd.lite` (image-to-image with a reference photo — never a physical model name);
   - `orientation` = `portrait` by default; use `square` or `landscape` only if the user explicitly asks for that frame;
   - `resolution` = `2K` by default; drop to `720P`/`1080P` if the user asks for a lighter render or the platform reports a downgrade in `param_adjustments`;
   - `name` = a short title such as `Bathtub Bath-Time Portrait`;
   - `entity_type` = `characters`.
7. **Verify and retry.** Compare the result with the uploaded photo using the checklist in `references/step-details.md`. If identity drifted, the towel/turban is missing, the scene is not a bathroom, or the anatomy is broken, retry once or twice with the corrective prompt fragments from `references/style-and-prompt.md`. Never present a failed generation as a success.
8. **Reply and guide.** Answer in one or two natural sentences describing the result (identity kept, towel + turban, warm bathtub). Do not paste long urls. If the roun