This surprisingly versatile SDXL LoRA is trained on a small dataset containing images of the remains of a huge fire at Philips Lighting Application Centre (LAC) Eindhoven, May 14th, 2023. The footage was taken two months later with a Canon G7X mkII point and shoot camera and is available on Huggingface.
Just drop one of the images into ComfyUI 🥳 or...
LoraLoader after the SDXL base model (strength usually at 1.0)
Both samplers DPM++ SDE
80 for the base, CFG 10
20 for the refiner, CFG 6
Standard SDXL resolutions
Many thanks to Stability.ai for SDXL, the wonders of ComfyUI, and TheLastBen for providing such a comprehensive SDXL LoRA training notebook 🙏