Workflow for Hunyuan video that can generate a small resolution video first very quickly, then upscales it with Hunyuan v2v when you find one you like. There is a third step for upscaling and video interpolation.
Version 1.5 uses the fast video lora to generate the first video in 7 steps, significantly increasing the speed of the first generation without compromising the second.
Version 1.6 uses a TeaCache sampler to increase generation speed by 1.6 and optionally by 2.1 with worse quality.
Version 1.7 adds Wavespeed, which has increased speed for me by about 15 %. To use it you will need to clone the wavespeed repo in custom nodes. Some wavespeed functionality requires installing triton, but if you only use the "Apply first block cache" node you may not need it.
If you already have a video you simply want to upscale, you can connect the muted load video node to the top left connection in the "Upscale and Interpolation" group and mute the previous 2.
This is just the application of some tips from this article with already available workflows.
This is not intended as tutorial on Hunyuan video, please check out the links above.
Description
FAQ
Comments (10)
Could you add an HD screencap of the workflow? I prefer to see the workflow before I go and download it. Thx <3
Sure thing! Should be up now.
Hey, good workflow. Is there a reason that you aren't doing (at least) the first pass using the fast model? You can output in 6 to 8 steps instead of 20. https://huggingface.co/FastVideo/FastHunyuan
I tried, but I couldn't get the fast model to consistently output decent quality videos on the second pass. Usually, I would get a pretty crappy first pass that the second wasn't able to correct unless I used very high denoise that removed the motion I wanted to keep. Eventually I decided I would lose less time by taking the generation speed hit of the bf16 model with more consistent quality.
I haven't tried using different models for each step, but you can avoid loading the 2 diffusion models by using the fast lora on 1 of the steps. Check out v1.5.
After each generation how do I reset to default and clear the previous previews and videos?
To be honest, I never thought to get rid of them before. There is a reload node option when right-clicking on nodes with previews, but I guess it will get tedious pretty quickly doing that manually. Check out custom nodes in the manager, maybe someone has made a simple button.
Thanks so much, it's a great workflow. BTW is there a way to upscale an existing video using [Load Video (Upload)] node?
Yes, I should probably add it to the description as well. Just above the "Interpolate and Upscale" group, I left a load video node to connect it to the top left blue image connection of the group. Just plug it in and mute the previous groups.
@bonetrousers Thanks for the quick reply. Is it possible to use the [Hunyuan Intermediate V2V] block instead of just using the Upscale model?
@mj_warrior Yes you can! You'll have to modify the workflow a bit. Get a "VAE encode" block and plug in the images from the video load block and the hunyuan vae. The samples output should be good to go to the "Upscale Latent By" block. You'll have to manually mute several nodes from the first group.