Retrained for Anima-Preview3

With more steps.

Result is very different. Some things are better, some seem worse.

Did not adjust the dataset, past issues still remain, namely:

- Blue tint leaning - 'blue' negative prompt as needed.

- If heads are too big - 'portrait' in negative prompt.

Enjoy!

I have another one of those assorted data loras to share.

It's not amazing, but there's not much action going on anyway. I guess, subpart lora is better than no lora.

This time it's animals!

Dataset was random, ranging from photos to illustrations. Includes mythical, real animals, "whimsical" animals, aquatic and a few horror animals. Does not include furries, this is just general 'animal' lora. But it does blend animals with humanoid shape quite well, because Anima.

It was a oneshot training, meaning it's messy. But it behaves nicely if one prompt engineers it a little.

A couple of glaring issues:

- Big heads. Because dataset did not differentiate portraits with tags. If happens - place 'portrait' in negative prompt.

- Very blue leaning. Likely because of aquatic data. Same as above - 'blue' in negative prompt.

Enjoy!