A Disco Elysium inspired LoRA using over a hundred images.
Model Information:
Primarily trained on portraits, this LoRA is fantastic at character generations, and mixes incredibly well with other styles. Includes version for SD1.5 and SDXL 1.0
How to use:
This LoRA has been trained on several varieties of prompt, but I've personally found best results using a [subject], [pose], [details], [style/lighting] style prompt. Prompts should be prefaced with "disco elysium" as an activation token.
Model works well with a variety of CFG scales, but lower (4.5-6) values are needed to bring out some of the artistic flourishes as seen in the game.
Description
Improved capabilities with landscapes and other non-portrait prompts. As the LoRA is still primarily trained on portraits, a lower model strength is still recommended.
FAQ
Comments (2)
Very interesting! nice work!
If you always get zoom on faces .. Just lower the weight.