This LORA model is trained using the technique described in https://www.crosslabs.org//blog/diffusion-with-offset-noise

Use it to produce beautiful high-key images that SD just wasn't capable of creating until now.

Sample images were generated using Illuminati Diffusion v1.0

The LORA was trained using Kohya's LORA Dreambooth script, on SD2.1-768 and SD1.5 with a dataset of 53 high-key, high-quality photographs.

Model weights can be used to affect the 'blown-out' lighting.

SD 1.5 version coming soon.