Type | |
Stats | 157 307 |
Reviews | (14) |
Published | Jul 18, 2024 |
Base Model | |
Training | Steps: 820 Epochs: 10 |
Usage Tips | Clip Skip: 1 |
Trigger Words | tokenframe |
Hash | AutoV2 BD6319A265 |
This is a LoRA designed to create token frames primarily for virtual tabletop role playing games such as D&D on Roll20 or Pathfinder on Foundry.
These are meant to have the white background removed via a photo editing program like Photoshop or GIMP and used to wrap around the edge of a rounded picture of your character.
Some of the images have also been tested with REMBG / BRIA and is largely successful with some minor edits required afterwards
The trigger word is "tokenframe" and it is recommend to also add "empty round frame, white background" to your prompt. This was designed to work with the base SDXL model and doesn't necessarily need the refiner though does work well with it. It seems to work well with multiple samplers but I would recommend either DPM++ 2M Karras or Euler. The LoRA was trained using 1024x1024 images and that is the supported resolution.
This is a retrained model of Game Avatar Token Frame done in conjunction with Sisimshow (hence borrowing the text from there) with additional imagery generated in Midjourney.
The LoRA works as is however It can sometimes be a little overzealous as 1.0, so perhaps try 0.7-1 depending on your generation prompt.