is this model for anime only?
good job bro for provide new model. is this model for anime picture generation only? or could we use for photography style?
it's not anime model, it does photography better.
But I tested anime only.
I will add photo sample soon.
thank you. and how to convert to NVFP4? i wonder can convert the text_encoders/ernie-image-prompt-enhancer.safetensors to NVFP4 format. Do we need expensive device, graphic card to convert NVFP4? thank you so much for provide us this NVFP4 format.
I have converted qwen3.5 using comfy-dit-quantizer.
You may refer that config.
https://github.com/bedovyy/comfy-dit-quantizer/blob/main/configs/te_qwen3.5-nvfp4mixed.json
You can check layers on huggingface (right button at the filename on list).
It doesn't need much resource but quality may not good as other quants.
Furthermore, nvfp4 clip is smaller but slower than bf16, so sometimes bf16 clip is better than quantized one. (especially, ComfyUI has VRAM optimization like dynamic vram.)
thank you for your information