Chen Mingyi

1) Regarding the disk size comparison between IQ4_XS and Q4_K_M, for a [model based on Qwen 3 32B](https://huggingface.co/bartowski/nvidia_Qwen3-Nemotron-32B-RLBFF-GGUF), IQ4_XS takes up 17.69GB compared to 18.77 GB for Q4_K_S and 19.76...

> 试了效果还是非常不错的，就是好像不支持多图，要是能保持多参考图人物的一致性将秒杀qwen edit I believe the model might be possible to take in multiple image inputs, because official pipeline implementation allows passing a list of images as input.

@llnancy The custom node you posted simply wraps around the official inference code, which itself simply inherits from the `diffusers` library. What we want is native integration of LongCat Image...