Spaces:
Sleeping
Sleeping
Commit History
-
Fail fast on hard Gemini quota instead of spamming every key
SuZeAIA 429 with 'limit: 0' means the model is unavailable on the plan for every key (e.g. free tier gives 0 image requests), so rotating through all keys just made N redundant calls and still failed. Detect that case and stop after one call. Surface a clear quota/billing message on the try-on task instead of a raw error.d47f0ef -
Switch try-on and garment analysis to external image APIs
SuZeAIReplace the local CatVTON+mask pipeline with a multi-provider image-generation service (Gemini / OpenAI gpt-image-1 / Replicate), selectable via TRYON_PROVIDER. Wardrobe item analysis now classifies garments via the Gemini Vision API instead of a local SegFormer model, so the backend needs no torch/transformers. - tryon_service: person + garment images -> generated try-on (sequential for multi-garment), provider chosen by env, clear errors when keys are missing - vision_service: Gemini garment classification (category/color/style_tags) with a simulated-metadata fallback - tasks: rewrite run_idm_vton_tryon -> run_tryon; drop CatVTON/SegFormer loaders - config/.env.example: add TRYON_PROVIDER, GEMINI_IMAGE_MODEL, GEMINI_VISION_MODEL, REPLICATE_*; drop CLOTHES_SEGMENTATION_MODEL, TRYON_MODEL and CUDA settings25d14ad
-
feat: default STORAGE_MODE to cloudinary and force real CatVTON execution on CPU
85a11ff -
feat: add CPU simulation fallback mode for tryon to prevent hangs/OOM on CPU spaces
6aa0990 -
perf: optimize CatVTON on CPU with float32 and 20 steps, increase frontend polling to 10 min
f222800
-
chore: move CatVTON inside backend and resolve paths dynamically without binary resources
f32f6d0
-
feat: integrate local CatVTON virtual try-on and secure env files
9a36956
-
Initialize production-grade FastAPI backend with models, routes, celery workers, and test suite
b18b789