Eigen-Banana-Qwen-Image-Edit LoRA trained on Apple's Pico-Banana-400k dataset with lightx2v/Qwen-Image-Lightning for 4-step inference 💨