v1.1 Update: Looks like I didn't test it enough because v1.0 looks pretty bad. Re-trained v1.1 after pruning some of the dataset and with more steps. This version should be less blobby and closer to the Mirror's Edge aesthetic.
XL LoRA trained on Mirror's Edge 1 screenshots. It's an update of an old dreambooth model but should be more flexible.
Trigger words: Mirror's Edge style, cityscape, abstract world, sterile, clean