This model was trained on 1300 photos of North American cities from a pedestrians view at night and golden hour, with many images in the rain. Originally a LoRA merged into Flux Dev.1 fp8. 15,000 steps on a A100 I used text encoding which caused the LoRA to fail in Invoke, and with Forge you can't use fp16 LoRAs with Euler. Merging to a checkpoint solves this. Please correct me if I'm wrong. Please comment with suggestions how I can improve, and if the results are what you are expecting I offer many interesting highly trained LoRAs for architecture at hallett-ai.com