Hunyuan Video I2V CFG Distilled version needed asap!
Thanks for this Kjai! Lots of workflow out there for cfg distilled version but no weights! lets go kjai cfg distilled version needed!
It is only available as cfg distilled version, I didn't name it that this time as there doesn't seem to be plans to release undistilled. Only undistilled HunyuanVideo model so far has been the Skyreels model.
okay i wonder why i am running into this issue then where i can use the hunyuan_video_720_cfgdistill_fp8_e4m3fn.safetens file from you with ease on my workflow for img2vid with the img2vid lora but not the hunyuan_video_I2V_fp8_e4m3fn.safetensors i am getting this error
From the Sampler Custom Advanced node: Given groups=1, weight of size [3072, 33, 1, 2, 2], expected input[1, 16, 16, 54, 38] to have 33 channels, but got 16 channels instead
weight of size [3072, 33, 1, 2, 2]
Means the model is I2V, the 2nd number refers to input channels and the I2V model has 16+16+1 for mask = 33. So that's right.
input[1, 16, 16, 54, 38]
Refers to the workflow not actually passing the image to the model, so this is a workflow or node version issue, is your ComfyUI up to date?
yes very updated anyways your example workflows for hunyuan imag2vid are woking fine though i will figure something out for my workflow but i am kind of disappointed by official hunyuan img2vid, skyreels and wan img2vid are far better it seems in terms of character consistency
turns out my comfy ui did not update through the comfy ui manager even after showing update successful had to manually update it through the update folder located inside comfyui portable folder by running the update bat file, everythings working as expected now, thanks for pointing out the ComfyUI update issue kjai!