HyVid - High Quality & Fast Speed Generation for 24GB Cards - Hunyuan Video Workflow — v1.0

Author notes from HyVid_24GB_FastGen_TheAIDoctor_V1.json · selected release

When using multiple loras, enable the HunyuanVideo Lora Block Edit node(select then ctrl-m), and make sure the blocks are connected to all of the Lora blocks inputs.

Author notes from HyVid_24GB_FastGen_TheAIDoctor_V1.json · selected release

Going to be honest, not sure exactly what this does but it's necessary for the Loras to function

Author notes from HyVid_24GB_FastGen_TheAIDoctor_V1.json · selected release

If you have tritron installed and configured you can use this to reduce VRAM. If you get this error because of your torch version: "torch.inductor error: Cannot create a file when that file already exists" You can edit a torch file to fix it: https://github.com/woct0rdho/triton-windows/issues/10 https://github.com/kijai/ComfyUI-HunyuanVideoWrapper/issues/189#issuecomment-2558649417

Author notes from HyVid_24GB_FastGen_TheAIDoctor_V1.json · selected release

Can use {option1|option2|option3} format to get a random option each time

Author notes from HyVid_24GB_FastGen_TheAIDoctor_V1.json · selected release

Upscales to fixed HD

Author notes from HyVid_24GB_FastGen_TheAIDoctor_V1.json · selected release

Frame interpolation, need to double the frame rate since you double the frames.

Author notes from HyVid_24GB_FastGen_TheAIDoctor_V1.json · selected release

Upscales by 4

Author notes from HyVid_24GB_FastGen_TheAIDoctor_V1.json · selected release

Enabling this node below uses the enhance a node to improve the video quality and style but it's not good if you are just testing the pure output of a lora. https://www.reddit.com/r/comfyui/comments/1hk26qc/hunyuanvideo_with_enhance_a_video/

Author notes from HyVid_24GB_FastGen_TheAIDoctor_V1.json · selected release

Requires the following Custom Node Packages Impact Pack - (For Wildcard Prompts) - https://github.com/ltdrdata/ComfyUI-Impact-Pack KJNodes - (For Get/Set Nodes) - https://github.com/kijai/ComfyUI-KJNodes HunyuanVideo Wrapper - (For Loading Hunyuan Models) - https://github.com/kijai/ComfyUI-HunyuanVideoWrapper VideoHelperSuite (For creating video output) - https://github.com/Kosinkadink/ComfyUI-VideoHelperSuite ComfyUI Frame Interpolation (For Frame Interpolation) - https://github.com/Fannovel16/ComfyUI-Frame-Interpolation Requires the following Models Hunyuan Video Safe Tensors - I like the BF16 best - https://civitai.com/models/1018217?modelVersionId=1141746 Hunyuan Video VAE - https://civitai.com/models/1018217?modelVersionId=1141674 Hunyuan Text Encoders - https://github.com/Tencent/HunyuanVideo/blob/main/ckpts/README.md#download-text-encoder Upscale and Frame Interpolation Models 4x_foolhardy_Remacri - https://huggingface.co/FacehugmanIII/4x_foolhardy_Remacri Film Net FP32 - https://civitai.com/models/1167575?modelVersionId=1314385

Author notes from HyVid_24GB_FastGen_TheAIDoctor_V1.json · selected release

Can use to lower VRAM usage if needed

Author notes from HyVid_24GB_FastGen_TheAIDoctor_V1.json · selected release

I designed this specifically around a RTX 4090 - 24GB VRAM running on Windows 10. I will not be able to provide reccomendations for other hardware or operating systems. I much prefer the outputs of the BF16 model of Hunyuan video, with the Triton Torch Compiling and TeaCache I get generations in 90-120 seconds (once the cache is warmed up) So far I found the best speeds and outputs at 416x736 resolution (or flipped), 121 frames (which at 24fps gives you 5 second clips) and steps between 20-30 (depending on your Loras) Most Lora's I've tested work well between .5 and 1. You may need to lower the strength on one or more of the Loras when mixing them. If you get blurry/grainy videos try lowering. Most of these discoveries have been trial and error, especially around the Triton Torch Compiling and TeaCache so I likely can't give much guidance outside of whats currently in the settings. Installing Triton on Windows was a nightmare - I used this guide: https://old.reddit.com/r/StableDiffusion/comments/1h7hunp/how_to_run_hunyuanvideo_on_a_single_24gb_vram_card/ You can find my Loras, models and workflows on my profile on Civitai: https://civitai.com/user/TheAIDoctor If you find this Workflow help please consider donating buzz so I can further my Lora Trainings and WorkFlow development. If you discover any improvements or issues please leave a comment on the post.

Author notes from HyVid_24GB_FastGen_TheAIDoctor_V1.json · selected release

Helps speed up Gen Times Significantly by caching the model. Mute node if cannot use.