Innovate Futures @ Benji
patreon
[Workflow] Mochi1 V2 With Unsampling txt2vid & img2vid (for Patreon Supporters)
🕑 Added 2024-11-11 13:00:09 +0000 UTC
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! :)
