Personal color analysis

Goal: a styling or commerce agent turns portrait photos into a defensible color verdict - measured, not vibes.

Step 0 - free quality precheck

POST /v1/portrait-precheck   free (10/hour/IP)
tells you BEFORE paying whether a face is detectable and the photo
quality is sufficient (lighting, color cast)

Step 1 - analysis

POST /v1/portrait-analysis   flat $0.05 for 1-3 photos
input:  {"images": [{"image_url": "..."}], "style": "feminine|masculine|neutral"}
output: measured colorimetry (CIELAB, ITA angle) + vision-model consensus,
        4-season verdict by published rules + all season_scores,
        structured palette (base/accent/metal/avoid + reasons), task advice

Step 2 - deliverable (optional)

POST /v1/style-report   $2.99 (10-pack $19.90 via /v1/report-pack)
finished PDF report (en/ja) an agent can hand to a human client

Cross-photo lighting consistency is checked; low-quality measurements are honestly downgraded to reference values (quality flags in the response). No identification; images are not stored. Calibration: 16-subject diverse-skin set - see /vision-calibration.