Innovate Futures @ Benji
patreon
Mochi 1 In ComfyUI Tutorial Guide To Generate AI Video Using Native And Wrapper Nodes
🕑 Added 2024-11-07 13:24:38 +0000 UTC
Video: https://youtu.be/mG1YqJK1p0g
Run DiT AI Video Locally on Your PC! In this game-changing tutorial, I'm revealing how to use the new Mochi One AI video model through ComfyUI's latest update. Finally, you can create professional AI videos on consumer-grade hardware! I'll show you exactly how to set up and optimize these models for Nvidia 4090 GPUs, breaking down the difference between FP8 and FP16 implementations.
------------------------------------------------------------------------------------
Running Mochi 1 Using ComfyUI Custom Node
ComfyUI-MochiWrapper
https://github.com/kijai/ComfyUI-MochiWrapper
https://huggingface.co/Kijai/Mochi_preview_comfy/tree/main
Another Method
------------------------------------------------------------------------------------
Native ComfyUI Running Mochi 1
***Make sure you update your ComfyUI before start.
Try the following steps to run the Mochi model immediately with a standard workflow.
1 - Update to the latest version of ComfyUI
2 - Download Mochi weights(the diffusion models) into models/diffusion_models folder
https://huggingface.co/Comfy-Org/mochi_preview_repackaged/tree/main/split_files/diffusion_models
3 - Make sure a text encoder is in your models/clip folder
https://huggingface.co/Comfy-Org/mochi_preview_repackaged/tree/main/split_files/text_encoders
4 - Download VAE to: ComfyUI/models/vae
https://huggingface.co/Comfy-Org/mochi_preview_repackaged/tree/main/split_files/vae
Therefore:
diffusion_models/mochi_preview_bf16.safetensors goes in: ComfyUI/models/diffusion_models/
text_encoders/t5xxl_fp16.safetensors goes in: ComfyUI/models/text_encoders/
(If lower VRAM download t5xxl_fp8 )
vae/mochi_vae.safetensors goes in: ComfyUI/models/vae/
Low RAM Solution Method 2 all-in-one checkpoint model
1- Download the packaged checkpoint to the models/checkpoint folder
https://huggingface.co/Comfy-Org/mochi_preview_repackaged/tree/main/all_in_one?ref=blog.comfy.org
2- Run a simplified video generation workflow as above.
Source :
https://blog.comfy.org/mochi-1/
https://comfyanonymous.github.io/ComfyUI_examples/mochi/
Comments
Benjamin Law
Update your comfyui, and try it.
aprisma
Hi. Doesnt matter if I use the all in one workflow with all in one model or the vae, text encode separated, I always get this message: KSampler meshgrid expects all tensors to have the same dtype Any idea?
