Kijai

/

LTX2.3_comfy

like 186

U can try in combo gemma nvfp4 uncensored https://huggingface.co/Sikaworld1990/gemma-3-12b-qat-abliterated-sikaworld-fp4-ltx2

I downloaded the Model from the Lightricks/LTX-2.3 Repo but I am getting the "RuntimeError: mat1 and mat2 shapes cannot be multiplied (13640x4096 and 2048x4096)"

I made it again.
https://huggingface.co/Bedovyy/LTX2.3_transformer_only_comfy

I downloaded the Model from the Lightricks/LTX-2.3 Repo but I am getting the "RuntimeError: mat1 and mat2 shapes cannot be multiplied (13640x4096 and 2048x4096)"

I am getting same error

I downloaded the Model from the Lightricks/LTX-2.3 Repo but I am getting the "RuntimeError: mat1 and mat2 shapes cannot be multiplied (13640x4096 and 2048x4096)"

更新你们的comfyui到最新版本。

Is anyone getting positive results from the NVFP4 model? I have it working, but the results are terrible. Its suppose to be similar to BF16, but faster. Something in the settings must be wrong, but I've tried A LOT of conditions.

I made it again.
https://huggingface.co/Bedovyy/LTX2.3_transformer_only_comfy

wow the nvfp4 is the only one that have bad fingers in your examples 😯

btw, why not creating mxfp4 quantization too?

I made it again.
https://huggingface.co/Bedovyy/LTX2.3_transformer_only_comfy

wow the nvfp4 is the only one that have bad fingers in your examples 😯

btw, why not creating mxfp4 quantization too?

Simple. ComfyUI doesn't support yet.