Skip to content

Implement NAG on all the models based on the Flux code.#12500

Merged
comfyanonymous merged 2 commits intomasterfrom
temp_pr
Feb 17, 2026
Merged

Implement NAG on all the models based on the Flux code.#12500
comfyanonymous merged 2 commits intomasterfrom
temp_pr

Conversation

@comfyanonymous
Copy link
Member

Use the Normalized Attention Guidance node.

Flux, Flux2, Klein, Chroma, Chroma radiance, Hunyuan Video, etc..

Use the Normalized Attention Guidance node.

Flux, Flux2, Klein, Chroma, Chroma radiance, Hunyuan Video, etc..
@comfyanonymous comfyanonymous merged commit 1892753 into master Feb 17, 2026
14 checks passed
@comfyanonymous comfyanonymous deleted the temp_pr branch February 17, 2026 04:30
@yamfun
Copy link

yamfun commented Feb 18, 2026

Seeing this, I tried adding a "Normalized Attention Guidance" between Load Model and CFGGuider of my existing Klein Edit WF and seems the result is the same? Hope there is a sample workflow for Klein Edit thanks.

@zwukong
Copy link

zwukong commented Feb 20, 2026

yeah, result is the same in Klein 9B Edit

@slikvik55
Copy link

Could we at least get a few lines in the changelog on how to use this!? I can't see how you can merge the negative prompt into the model when the NAG node only has a model input.

@Castelmatto
Copy link

Castelmatto commented Feb 23, 2026

This does absolutely nothing compared to the official custom node, not on Chroma, Chroma Radiance, Klein or Flux2. Setting all values to their maximum, the outputted image remains the exact same while it should be completely broken. Something is wrong on how it has been implemented. Why it has been merged without proper testing first? Same thing happened with the Image Comparer node which is also broken.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

5 participants