Can i use my already downloaded Wan Models?

[AndreiSavchuk]

or i need automatic download it again?

[jaimitoes]

if your models are place in the correct path /models/diffusion_models/ you can use all your current models.

[zappazack]

Is it possible to use the Quant models f.e Q8?

[jaimitoes]

@zappazack No, gguf are not allowed, you can use the trained Lora with a gguf but not train a Lora with it.

[unjadded]

I placed Wan14BT2VFusioniX_fp16.safetensors in diffusion folder and for some reason the node cannot see it...can't start training. Tried to switch it to so many directories, not sure what's wrong...

[ptitjes]

@jaimitoes same here. I put my wan2.2_t2v_high_noise_14B_fp16.safetensors and wan2.2_t2v_low_noise_14B_fp16.safetensors in my models/diffusion_models/ directory as usual to use them as DiTs. But the dit_name combo box of the "Musubi LoRa Trainer" node stays empty. (I run ComfyUI in a Linux docker container.)

[jaimitoes]

@ptitjes This only trains Wan 2.1 models, the project was created before the launch of Wan 2.2, so you are not able to train Wan2.2 models. @unjadded if you have the dit model in the folder "models/diffusion_models/ it should work, if not, maybe you have a path problem issue or missing some modules.

[PaTiToMaSteR]

Is there any posibility to train Wan 2.2? Or anybody recommends another project?