OmniGen2 plots

Default workflow from https://github.com/comfyanonymous/ComfyUI/pull/8669

I forgot to write it on the plots but the "default" text cfg is 5.0 (the middle value)

Very limited takeaways so far:

  • Seeds make a huge difference in not just composition but input image adherence
  • Imperative prompt seems to have slightly better input image adherence, perhaps at the cost of more distortions
  • To edit images with simple prompts without altering them too much, lowering text cfg to 3 or even lower seems like a good move
  • Increasing image cfg also increases input image adherence in a different way, seems to deteriorate outputs a little


Declarative prompt

"She is wearing a red beret."

x: text cfg | y: image cfg

x: seed | y : image cfg

x: seed | y: text cfg


Imperative prompt

"Give her a red beret."

x: text cfg | y: image cfg

x: seed | y: image cfg

x: seed | y: text cfg

Edit
Pub: 26 Jun 2025 15:59 UTC
Edit: 26 Jun 2025 17:35 UTC
Views: 130