Trained on sd 1.5, recommended strength between 0.6 and 0.8. Personally I think 0.8 gives the best results. For the training process I used 20 images 20 repeats 10 epochs, max steps 2000. For the generation process, most of the example pictures I did a standard txt2img on 512*768, with hires on 1.5 scale and swinir as upscaler. Denoising between 0.45 and 0.65.