wangkanai

/

wan22-fp8-i2v

like 1

I'm following the instructions for "Image-to-Video Generation (FP8)", but that only loads the high noise weights with

pipe.unet.from_single_file(
    "E:/huggingface/wan22-fp8-i2v/diffusion_models/wan/wan22-i2v-14b-fp8-high-scaled.safetensors"
)

What attribute do I replace (instead of unet) for the low noise model?

When I look at other Wan docs, they use WanImageToVideoPipeline and WanTransformer3DModel to load transformer for high noise and transformer_2 for low noise but that gives me other errors entirely...