Version 2.0

Flexible across all sorts of media, except photos, because there was no data to tell it what a real avali would look like. (Go figure.) Prompting tips:

• Really, really likes Gleipnir v0.3.

• Use at weight 0.8 to 1.0, depending on art style.

• Works much better at SDXL resolutions like 1024x1024 even if your model supports lower.

• If something like the eyes doesn't come out right, inpainting at a higher resolution can make it much better.

• Use CFG Rescale to play with color balance if it gets weird. I usually find myself generating at 0.5 in most styles just because I like the less saturated look.

• Try making avali-shaped food, it's fun!

Version 1.0

This version is less flexible, but has some SD 1.5 gens to tell it what a photorealistic avali might look like. Prompting tips:

• Most of my captions used "male avali" or "female avali", so those'll probably be stronger than "avali" alone.

• Body colors were tagged in the format "brown feathers with white countershading and orange borders".

• Colors in the images can get way oversaturated. I had good luck using CFG Rescale to correct that.

描述:

Trained on 50 images with lazy auto-captioned booru tags. Transfers better to various media, but can't do photorealism.

训练词语: avali