Person → Animal Silhouette (典雅动物裁切)

Transform a person's photo into a tasteful animal-shaped composition: the subject is
recognized, their pose and expression are read, the photo is composited into an
animal silhouette that matches them, and an elegant color scheme is built
from the chosen animal's aesthetic "effect" (效应).

When To Use

Trigger this skill when the user:
- Uploads or references a photo of a person and asks to crop / clip / mask it into an animal shape.
- Asks to turn a portrait into an animal silhouette / outline / shape.
- Requests "典雅" (elegant / refined) animal-themed portrait art with a matching color scheme.
- Uses any of these trigger phrases (Chinese or English):

中文触发词
- 把这张人像裁成动物形状 / 裁成动物轮廓 / 裁切到动物轮廓
- 人物变动物 / 人像动物化 / 照片动物化
- 根据动作表情匹配动物并把照片裁成动物
- 给我做一张典雅的动物剪影人像 / 动物形状配色肖像
- 识别照片里的人，按他的姿势配一个动物，再配典雅色

English trigger phrases
- "Crop my photo into an animal silhouette"
- "Turn this portrait into an animal shape"
- "Make an elegant animal-shaped portrait from this person"
- "Match this person to an animal and recolor it elegantly"

If the request is only about generic photo styling (not animal-shaped), or the image
has no clear human subject, say so and either ask for a person photo or decline.

Workflow

Follow these phases in order. Phases 1–3 are analysis (no generation); Phase 4
generates; Phase 5 refines.

Phase 0 — Intake

1. Obtain the source image. If none is provided, ask the user to upload a single
   human-subject photo (a clear, well-lit portrait or full-body shot works best).
2. The skill always renders with Mode A (AI compositing) — the most elegant
   path, using the ImageGen tool in image-to-image mode. There is no offline
   clip mode; every output is an AI-composited animal-shaped portrait.

Phase 1 — Person Analysis (vision)

Use native multimodal vision on the source image and produce a structured analysis
(keep it short, one line per field):

• pose: standing / sitting / crouching / jumping / arms-raised / side-profile /
  lying / dancing / running / reaching
• action: what the body is doing (e.g., reaching out, hugging self, playing instrument)
• expression: smile / laugh / calm / serious / surprised / thoughtful / playful
• mood: serene / energetic / mysterious / warm / regal / whimsical
• silhouette_features: notable outline traits (extended limbs, long hair, hat,
  long coat, ponytail, glasses)

This structured reading drives the animal match. Do not skip it — vague analysis
produces a weak match.

Phase 2 — Animal Matching

Consult references/animal_mapping.md. Match on two axes and state the result
to the user in one sentence:

1. Pose-silhouette axis — which animal's body outline best contains or echoes
   the person's pose (e.g., arms-raised → bird/crane with spread wings;
   crouching → cat; seated-calm → rabbit/deer; side-profile → horse/fox;
   jumping/leaping → deer/gazelle; dancing → butterfly/peacock).
2. Temperament axis — which animal's symbolic character matches the
   expression/mood (calm → deer/crane; playful → fox/rabbit; regal → peacock/lion;
   mysterious → cat/owl; energetic → horse/dolphin).

Conflict rule: prefer the pose-led match for shape fidelity; let temperament
refine nuance and the palette. If the two axes point to very different animals and
the user clearly cares, ask them to choose, or present both options.

Output: animal = <name> + a one-line rationale (e.g., "Arms raised + serene →
crane: spread-wing silhouette fits the pose, crane's calm symbolism fits the mood").

Phase 3 — Elegant Palette

Consult references/elegant_palettes.md, keyed by the chosen animal. Each entry
provides:
- a 4–6 color 典雅 palette (hex codes),
- a named mood, and
- an aesthetic direction (e.g., 水墨留白 / 莫兰迪 / 珐琅彩 / 秋日暖调).

"效应" interpretation: the palette is derived from the aesthetic effect the animal
evokes — crane → 清雅留白的水墨感; peacock → 宝蓝描金的华贵感; fox → 秋日暖橘的灵动感.
Show the palette swatches (hex) to the user and let them approve or tweak
(e.g., "make it cooler / warmer / more muted").

Phase 4 — Generation

Generation (AI compositing):
Use the ImageGen tool in image-to-image mode with the source photo as the
reference image. Build the prompt from this template (fill <…> from Phases 1–3):

Composite a portrait photograph into a clean, full-bleed <animal> silhouette.
- Keep the subject's facial likeness and clothing texture visible as the "fill"
  inside the <animal> outline; preserve recognizability.
- The <animal> shape must be a crisp, continuous silhouette that contains the figure.
- Background: an elegant <aesthetic_direction> field using this palette: <hex_list>.
- Mood: <mood>. Refined, tasteful, gallery-quality, "典雅" (elegant) sensibility:
  low-to-mid saturation, generous negative space, sophisticated, not cartoonish.
- Fine outline, soft grain, editorial portrait lighting.

Keep the prompt tight; avoid contradictory style words. If the first result loses
likeness or the shape is weak, regenerate with stronger "preserve facial likeness"
and "crisp <animal> silhouette containing the whole figure" emphasis.

Phase 5 — Present & Refine

1. Present the result via present_files (image/preview).
2. Offer refinements:
   - swap animal (re-run Phase 2),
   - swap/tweak palette (re-run Phase 3),
   - adjust elegance (more/less saturation, more negative space),
   - add a frame, caption, or the animal's name in elegant typography,
   - regenerate Mode A with a different seed for a fresh composition.

Resources

references/

• animal_mapping.md — pose/expression → animal matching tables (two-axis).
• elegant_palettes.md — animal → 典雅 color scheme (hex + mood + direction).

Notes & Pitfalls

• Best results with a single, clearly separated human subject; group photos or
  busy backgrounds weaken the match and the silhouette.
• Mode A preserves likeness but the silhouette is AI-interpreted (may soften edges);
  this is intentional for the "典雅" aesthetic and gives the most refined result.
• Always show the chosen animal + palette to the user before the final generate,
  so the "match" feels intentional rather than random.
• Keep palettes "典雅": favor muted, harmonious, low-saturation tones; avoid neon
  or high-contrast clashing colors unless the user asks.