IMPORTANT REQUEST: Please add "banana qwen-image-edit" svdquant versions !
Hello, I've been following your work closely. Your optimizations for consumers with limited VRAM capacity are fantastic and unique in this field. I congratulate you on this and hope you continue to do so. I have a request. It would be very helpful if you could also use SVDQ Quant for the Nano-Banana model (https://huggingface.co/eigen-ai-labs/eigen-banana-qwen-image-edit), which has recently made a difference in image rendering quality and consistency. If possible, we, as nunchaku followers, would be very happy if you released a lightweight 4/8-step quantized model like the NF4. I would appreciate your consideration and evaluation.
This is just a lora. Just load it with the experimental lora loader no? as far as i know this also has some issues and isn't a consistent improvement