Pet Selfie · Open Skill 공개 스킬
Pet Selfie
Pet Selfie
Internal
0 1 1
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/3656543/2026-10-14/crjyh687.jpg
# Pet Selfie
Generate **pet selfie** images: the pet appears to be taking its own photo (selfie stick, paw holding a smartphone, bathroom mirror, car dashboard cam vibe, etc.). Preserve the pet’s identity when a reference photo is provided.
There is **no structured project state**—progress lives in the conversation and canvas assets.
## References (optional)
Read **only when** the user asks for a specific selfie vibe or you need phrasing help:
| Topic | File | When to read |
|--------|------|----------------|
| Selfie moods & prompt snippets | `references/selfie-styles.md` | User wants mirror/car/cozy/holiday/etc. styling |
## When to use
- User asks for a pet selfie, “my cat/dog taking a selfie,” funny/cute pet phone photo, or similar.
- User uploaded a **pet photo** and wants a selfie version of **that** pet.
- User has **no photo** and describes the pet in text (breed, color, expression, setting).
- User wants **another variant** (different pose, outfit, location, or selfie style) of a pet already in the thread.
## Out of scope
- Non-pet subjects as the main focus, human-only selfies, or explicit/NSFW content (`.hd*` image channels do not support NSFW—decline politely).
- Full short films or multi-shot video pipelines (use a movie/video skill instead).
## Workflow
### 1. Confirm intent (brief)
Extract or ask only what is missing:
| Input | Action |
|--------|--------|
| **Reference image** (user upload or canvas pet photo) | Image-to-image path (Section 2) |
| **No reference** | Text-to-image path (Section 3) |
| **Selfie style** | Default: smartphone front camera, slight wide-angle distortion, warm natural light, playful expression. Honor user specifics (mirror, beach, cosplay, etc.). |
| **Orientation** | Default `square` (classic social selfie). Use `portrait` if user wants vertical/stories; `landscape` if they ask for horizontal. |
| **Count** | Default **1** image per request unless user asks for multiple (then run separate calls; `batch_size` is always 1). |
Use `write_todos` only when the user asks for several distinct selfies or a small set of variations.
### 2. Image-to-image (user provided a pet photo)
1. **Understand the pet (recommended):** Call `recognize_image` on the reference URL with a short prompt such as: “Describe species, breed if visible, fur/coat colors and patterns, face markings, eye color, and any accessories. One concise paragraph.” Use this to keep prompts accurate—do **not** replace the reference image.
2. **Reference URL:** Put the user’s pet image in `generate_image` → `input_urls` as a one-element list. Prefer the upload/canvas `asset_id` or the platform URI from context; do not invent URLs.
3. **Model channel (required—pick exactly one):**
| Channel | When to use |
|---------|-------------|
| `i2i.hd` | **Default** when identity fidelity matters (same pet as the upload). |
| `i2i.hd.lite` | User wants a **quick draft**, lower cost, or “good enough” preview before a final HD pass. |
| `i2i.hd.gpt` | User wants a **more stylized or creative** selfie (stronger reinterpretation) while still anchored to the reference; or prior `i2i.hd` result felt too stiff and they asked for a fresher take. |
Do **not** use physical model names in `model=`. Do **not** use `t2i*` when `input_urls` is set.
4. **Prompt (English):** Combine identity cues from `recognize_image` with selfie staging, for example:
> Same [species/breed] as the reference, [colors/markings]. The pet is taking a selfie: [front camera / mirror / paw holding phone], [expression], [background]. Photorealistic pet photography, sharp eyes, natural fur detail, selfie framing with slight wide-angle feel, no human face as the subject.
Stress **same individual pet** as the reference; avoid changing species or coat pattern unless the user explicitly asks.
5. **Call:**
```text
generate_image(
prompt=<English selfie prompt>,
input_urls=[<pet reference>],
model='<i2i.hd | i2i.hd.lite |