I've got a "new" comparison of Kohya and AI Toolkit #SaveKohya

With the message attached to Kohya and its current stability issues I've been mentioning the two LoRAs I had trained on both as comparisons. This one for two characters from a random parody PSA I decided to try on both from the start as a comparison and to test how it worked with more limited data. This one for one of the most infamously cheap bosses in fighting game history I decided to try doing an AI Toolkit version of after the original had been published for a while as a comparison since it was new enough I wouldn't need to change the data and used weird type of data (pixels) for most of its data.

Turns out I actually had a third comparison LoRA already made because it came out so poorly I totally forgot I had it! This clothing LoRA was made on AI Toolkit first, and that version is now published here as a comparison. I think it was made by mistake (AI Toolkit is default) and I rolled with it by excusing the double training as a test, but I'm not sure. Whatever the case, I didn't publish it here because it just did not meet my standards at all (very high rate of inexplicable issues during testing for the sample gallery) while the Kohya version worked almost perfectly (and its relatively minor issues are understandable problems from the suboptimal training data options I was working with). (incidentally I actually have a second clothing LoRA I didn't publish at all here because it's broken during initial tests and I missed the holiday it was supposed to launch on as a result, but I have an idea on how to fix it and will try when Kohya is working again.)

Also I unexpectedly came into some buzz and thus posted a bounty.

Pic unrelated.