OmniGen in ComfyUI for Consistent Characters
Added 2024-11-10 21:54:27 +0000 UTCVideo - https://youtu.be/mwiz1PNDWGg
OmniGen is a multimodal model which can follow instructions in English. No need for IPAdapters, Depth or Canny controlNets, etc as OmniGen does it all. Ask it questions about the picture and get it to change any aspect you like!
One thing OmniGen does really well is create consistent characters, and even better still you can easily have up to three different ones in any one image!
Example image showing the standard workflow result:

GitHub Links
Original - https://github.com/VectorSpaceLab/OmniGen
ComfyUI Node - https://github.com/1038lab/ComfyUI-OmniGen
The ComfyUI node does everything for you, including downloading the OmniGen model files. Install using ComfyUI Manager as normal.
Workflows
As usual, the workflows used in the video are attached to this post. One is just the standard example, with the second including an Ultimate Upscale option using Flux and Detail Daemon.
Update!
The author has made loads of updates to the node recently, plus I noticed the custom ultimate upscale actually needs both custom option before it will actually do the custom stuff, so it's back to the standard UU with the v1 example.
The OmniGen node now also has performance related options for optimised performance, VRAM or a balanced mode. Included as a performance option is an FP8 version too :)
Another addition is the selection of preset prompts, though I still prefer making my own.
Comments
It's best with 12+GB VRAM, but will work with less
Nerdy Rodent
2024-11-12 07:39:42 +0000 UTCWhat sort of VRAM is needed to run OmniGen "OK" or just acceptable?
Chris & Tracy Mathews
2024-11-11 14:42:40 +0000 UTC