V1.0 DETAILS
I wanted the V1.0 Lora not to affect the style—only to change the lighting. It was trained with different styles to achieve this, so now this version simply makes the scene darker, minimally affecting the style. I also wanted the light sources to work better, which means the darkness isn’t as deep since the lights stand out more—but it paves the way for more cinematic generations. You can always revert to the previous version if you like.
Keyword: The Lora itself doesn’t have a keyword, but you can use dark, darkness, dark background to intensify the effect.
Training: It was trained on 130 images that were upscaled, inpainted, and batch-edited to adjust gamma and even out scene contrast, plus a few manually fine-tuned.
Issues: I noticed the Lora’s effect diminishes when too many tags are used, so keep it simple. If that’s not possible, add some extra tags. Also, using distinct light sources like candlelight can help improve the effect.
