Portrait Trend Poster

Identity-preserving youth-culture, fashion, music, idol, and editorial portrait posters from a required uploaded portrait. Built for AI image agents (Codex, Claude, etc.) that compose production-grade generation prompts.

The parent agent decides which to invoke based on user keywords and consent.

Any of these in the user prompt routes to glam mode (after consent check):

明星级, 明星海报, glam, celebrity, magazine cover, 棚拍大片, fashion cover, editorial portrait

Explicit mode: glam always wins. Explicit mode: preserve always overrides glam keywords.

Glam stages 2 (styling) and 3 (makeup) require explicit opt-in. Stages 1 (source prep) and 4 (style overlay) are on by default. Without consent, glam mode stops and asks for acknowledgment.

glam.consent: yes
glam.wardrobe: on
glam.makeup: off

Hard refusals (no generation, stops):

• non-consented glam
• identity replacement request
• ethnicity alteration
• celebrity impersonation
• facial anatomy change
• signature accessory invention

portrait-trend-poster/
├── SKILL.md                       # v1 entry point
├── agents/openai.yaml             # v1 invocation metadata
├── design-system/                 # 6 catalogs (colors, typography, ...)
├── evals/                         # 12 v1 behavioral cases
├── scripts/                       # v1 validators
├── examples/                      # 4 v1 recipe + render pairs
└── glam/                          # v2.0 subskill
    ├── SKILL.md
    ├── agents/openai.yaml
    ├── design-system/
    │   ├── stylings.json          # wardrobes, lightings, environments, makeup
    │   └── glam-consent.json      # stages, keywords, hard refusals
    ├── evals/                     # 8 glam cases
    ├── scripts/                   # glam validators
    └── examples/                  # 4 glam recipes

python portrait-trend-poster/scripts/validate_design_system.py
python portrait-trend-poster/scripts/validate_evals.py
python portrait-trend-poster/glam/scripts/validate_glam.py
python portrait-trend-poster/glam/scripts/validate_glam_evals.py
python -m unittest discover -s tests -v

All four scripts and the 63-test worktree-root suite must pass. Tests enforce byte-identical v1 SKILL.md, byte-reproducible release archive, and locked catalog contracts.

dist/portrait-trend-poster-v0.1.zip is the byte-reproducible archive built from git archive --format=zip --mtime=@1798912800 HEAD portrait-trend-poster with SOURCE_DATE_EPOCH=1798912800.

The skill produces prompts, not images. Pair it with an image-generation model (GPT-Image, Flux, Midjourney, Stable Diffusion, Codex ImageGen) to render the final raster.

v2.0. Glam is best-effort: image-model faithfulness for wardrobe / environment / makeup swap is not guaranteed. Identity preservation is enforced by must_not_glam in every glam eval case.