Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[BUG] Flux Lora training - sample image #2781

Closed
kalle07 opened this issue Sep 5, 2024 · 2 comments
Closed

[BUG] Flux Lora training - sample image #2781

kalle07 opened this issue Sep 5, 2024 · 2 comments

Comments

@kalle07
Copy link

kalle07 commented Sep 5, 2024

seems that if i generate sample images while the training, the lora is not used
so it works it generates images with the prompt i wrote but the image always the same if i take the same seed
if i use the lora seperate in compfy or webforge it works

@TheIronDev
Copy link

TheIronDev commented Sep 5, 2024

I am finding that the Lora is being used for me.

I pulled in the latest code from last night, and saw that the lora is used, but the sample images were not as good as what I saw in ai-toolkit.

If you're like me and used the config that SECourses suggested, the learning rate might be too low. I'm running another test but with a higher learning rate to see if the samples converge faster.

My Samples Screenshot:

Epoch 10

Screenshot 2024-09-05 075518

Epoch 200

Screenshot 2024-09-05 075530

Another thought, is maybe Guidance_Scale is off? I didn't see a setting in Kohya..

@kalle07
Copy link
Author

kalle07 commented Sep 5, 2024

i dont have the newest josn but
learning is 0.0005 (original was 0.00005)
also the image is generated two times faster than in webfroge_ui (same resolution)
same image size and no change after 10% after 20% and so on

@kalle07 kalle07 closed this as completed Sep 19, 2024
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

No branches or pull requests

2 participants