[Workflow] Mochi1 V2 With Unsampling txt2vid & img2vid (for Patreon Supporters) (Patreon)

Video : https://youtu.be/TK4b7uhHgkk

Related Post and basic workflow: https://www.patreon.com/posts/115743706

This workflow are using full features of both ComfyUI-MochiWrapper and ComfyUI-MochiEdit.I will add more "how to" in the description below for this workflow.

To use this workflow.

First, identify which method are you going for. txt2vid or img2vid?

click the img2vid to True if you want to start with img2vid

for the Groups Bypasser, enable Img2vid Generation, and disable txt2vid Generation

Then set the number of seconds in the blue box

Lastly , do your prompting

Denosing for Img2vid

Because this is only using Image Encoding for img2vid method, just like img2img concept.

So set the denoise between 0.5-0.6 are preforming good for this case. If you set 0.2-0.3 , there will be no movement at all.

Hopefully , Genmo AI will release the official Img2vid model weights, so we can really animate motion using image.

VAE Tiling must be true

I don't care what graphic card you use locally; it will be OOM error if you turn it off.

Or you can get a H100 then you can using this AI Model generate video without VAE Tiling.

Model Loader - Use whichever model version suite your PC.

High End GPU , Clip FP 16 is running smooth

Otherwise use FP8 

Steps for Sampling

1st sampling set it 20-30 steps is good enough. Don't go crazy on it.

Resampling Group , number of steps for scheduling , better to use 50+ here.

Any one who have question / needs assistance, please feel free to join communicate in over there.

Discord Private Group. https://discord.com/invite/kKcbdDJZmD

Have fun! :)