Is it possible to cancel caching of the final result to to generate with the same parameters? #211
-
|
When generating images with exactly the same input data, there is still a side effect that makes them different:
I would like to use this as sometimes it can solve some picture problems, but because of caching, you have to change prompt, and then return Template: https://comfyanonymous.github.io/ComfyUI_examples/area_composition/square_area_for_2_subjects.png |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment 4 replies
-
|
if you add one or more spaces to your text prompt it won't actually change the tokens but it will rerun it. The side effect is probably due to the SDE sampler and something I will need to fix. |
Beta Was this translation helpful? Give feedback.
-
|
Thank you, hope for improvement, good luck! But such the side effect is sometimes convenient, as I showed above, I would like to use it or something similar |
Beta Was this translation helpful? Give feedback.
-
|
@comfyanonymous This behavior, as I understand it, is due to |
Beta Was this translation helpful? Give feedback.
-
|
starting from 0.0.19 xformers should be deterministic. |
Beta Was this translation helpful? Give feedback.
-
|
regarding SDE and it's deterministic abilities /tmp/ComfyUI/auto/lib/python3.10/site-packages/torchsde/_brownian/brownian_interval.py:594: UserWarning: Should have tb<
=t1 but got tb=5.516684055328369 and t1=5.516684.
warnings.warn(f"Should have {tb_name}<=t1 but got {tb_name}={tb} and t1={self._end}.")maybe related |
Beta Was this translation helpful? Give feedback.
if you add one or more spaces to your text prompt it won't actually change the tokens but it will rerun it.
The side effect is probably due to the SDE sampler and something I will need to fix.