Suggestion: add fp8 and GGUF quants support for i2v (wan/hunyuan)

[mpdmtr]

Firstly I would like to appreciate the creator for a great tool.

As a casual user I need a support for pre-quantized FP8 and GGUF models. I have a weak machine and i2v cause crash even with turned on swapping and built-in FP8 functions. Also this support could help save a disk space for those who doesn't want to download 28-32 GB for checkpoint.

[maybleMyers]

Have you tried wan2gp https://github.com/deepbeepmeep/Wan2GP? It is already optimized for low vram/ram machines. I do want to add gguf support at some time and fp8 should already work with some models.
What model specifically would you like to use with gguf?

[mpdmtr]

Have you tried wan2gp https://github.com/deepbeepmeep/Wan2GP?

No, didn't hear about it. Thanks for advice, will also try.

fp8 should already work with some models.<

With wan i2v fp8 it doesn't work, gives an assertion error (I tried every variant of i2v presented in program's listbox). When I tick a checkbox "FP8" the system gives me a message that the model is already quantized and nothing happens.

What model specifically would you like to use with gguf?

I'm the most interested in wan i2v.

Upd:
Tried wan2gp. TBH I don't like cases when the program tries to download files I don't need. The same experience I had with Framepack, Also a possibilty of swap control also attracted me in H1111.

[maybleMyers]

Pre quantized Fp8 should load fine now in the WanX-i2v tab. Fp8 scaled is better quality though.

[maybleMyers]

To use fp8 e4m3fn model select a e4m3fn dit model and select use fp8. To use fp8 scaled select a full fp16 model then click Use fp8 scaled, it will scale it on generation.

[mpdmtr]

Thanks, now it works.