Oh good. I didn't know the Patreon version had fixes. I will have to join that. I've been doing a lot of playing with this to see how well I can control it, but I've run into another bug. This issue, is that I'm getting different results with the same seed/settings. I did three runs under the same seed, but different total images. I used the following settings on all:
Image mode: Resample
Prompt: Gambit vs Sentinel
no origin or target image
Hardcoded BQGan and Clip models
Use half-precision
Seed: 2
Generate Extra Images: 0
Resolution X:336 Y:496
Write Image every X frames: 50
Frames before stopping: various
And here are the images, which all have the same initial image, but are using the same seed to get predictable results (I was trying to go further on an idea when I saw it was just starting to get interesting). They are all image number 600:
851 limit:
Image limit 1251:
Image limit 901:
I'm using the same seed, so I should be getting the same results, but it's different each time. This makes it impossible to do some tests with some low image numbers, then extend on those tests when I find some good ones worth exploring. I have to always commit to a very high image count just in case I happen to get a good one.
Maybe this is fixed too? Either way, I'm joining the patreon. Thanks again.
