You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Even with the same settings & seed img2img always has significant randomness & changes in output. Occasionally output is the same as expected.
Problem
This behavior makes it extremely difficult to experiment with settings and see their effects. As it's impossible to attribute the changes to the settings or inherent randomness in the output.
Expectations
img2img generates consistently the same output with same input.
How to Replicate
After 2 days of experiments I think I traced it back to the model being loaded. If the img2img is generated on the same instance with the following message in the log .. weights already loaded then the output is consistent with images generated prior.
If the im2img happens on 2 different model instances with the message Ensuring enough disk space... Loading fine-tuned model in the log - the outputs are different.
Example:
*the differences are more pronounced at higher resolutions like 1024
Missing Leg
Big Face, 3 Ears
The text was updated successfully, but these errors were encountered:
Current Behavior
Even with the same settings & seed img2img always has significant randomness & changes in output. Occasionally output is the same as expected.
Problem
This behavior makes it extremely difficult to experiment with settings and see their effects. As it's impossible to attribute the changes to the settings or inherent randomness in the output.
Expectations
img2img generates consistently the same output with same input.
How to Replicate
After 2 days of experiments I think I traced it back to the model being loaded. If the img2img is generated on the same instance with the following message in the log
.. weights already loaded
then the output is consistent with images generated prior.If the im2img happens on 2 different model instances with the message
Ensuring enough disk space... Loading fine-tuned model
in the log - the outputs are different.Example:
*the differences are more pronounced at higher resolutions like 1024
The text was updated successfully, but these errors were encountered: