Qwen-Image configured for transparency and performance. It uses the official HuggingFace model for provenance, BF16 for numerical stability, and includes an optional Lightning LoRA. ⚠️ Each manual test = full cold boot cost!