Tested on Pony v6.

This model is a little unstable due to a small dataset and my limited training experience. Either that or I'm just spoiled on Illustrious at this point.

Works best with realistic and portrait tags. Also trained on the tags traditional media, watercolor (medium), and lineart.

I recommend a strength of 0.6-0.8

Recommended parameters:

• Sampler: DPM++ 2M Karras, some success with Euler a and DPM++ 2M SDE

• Steps: 25-40

• CFG: 4-6

• check samples for prompts

Example images utilized inpainting and SD Upscale.