Rainbow Brite, the original, in upscaled glory...

0.6, (0.5B-40)

Identical in settings to 0.5B-20 the fully cooked with 40 repeats version is the best one yet. At full strength it forces 80s style and dial it back to like 0.7 to get the appearance of the old set but with way better prompt response. Even with heavy lora loading the appearance is way more stable.

0.5B-20 is underbaked but it makes some interesting effects in this form so I'm going to give it its own section for people to remix down the road. The fully cooked version will be up soon.

version 0.5B

New image set, went from a handful to 79 but it's over training some concepts now while doing the boots way better. I concentrated too much on one thing but the important thing is that DATA happened. I know what happens now with both overcooking and undercooking as well as I can see how it's learning and what to add or remove to get what I'm after.

The 5B-20 model is underbaked I'll post the 40 reps version when it's done in a couple hours.

Version 0.1

I'm actually shocked this is working, there isn't alot of high res images available so this is upscaled to heck and based on the only movie I could find that's at least 720.

I messed up the training language, trying for rear view goes to the star stealer dungeon.

Beta: watch it for tweaks.

Version 0.5, OH right I have carpal tunnel, i forget from time to time, but DAMN, the intent is to make a lora good enough to frame by frame regenerate an old cartoon, damn near part of it done.