...
Run the Fashion-AI promo-photo pipeline from liangdabiao/Fashion-AI: feed a flat-lay product photo, hybrid-search a Milvus vault of bestseller references (dense image embeddings + TF-IDF sparse + sales filter, fused with RRF), analyze their shared scene/lighting/pose with Qwen3.5 VL, and generate a model-on-product promo shot with Nano Banana 2 / Gemini 3.1 Flash Image or OpenAI gpt-5 image variants through OpenRouter. Covers setup, search, full-pipeline generate, model swapping, aspect ratio, sales threshold, and prompt tuning for cross-border e-commerce sellers.
Erstellt von: songlin she · GitHub