This is a fairly flexible LoRa. If you are trying to do something extra stylized, though, it may help to change the weight to 0.75 if the results are still trying to be more photorealistic.

For all of my results, I am using the StyleSelectorXL extension in Auto1111. For this LoRa, I really like the 3D Model, Anime, Fantasy Art, Neon Punk, Fighting Game, Techwear Fashion, and Neon Noir styles. If you haven't worked with this yet, it automatically changes your prompt with the necessary tokens, but you can just toggle a different style to change without having to delete the tokens.

My typical workflow:

1. SDXL Base 0.9 checkpoint

   1. SDXL Offset Example LoRa 1.0

   2. Euler

   3. 20-30 steps

   4. 7-9 CFG

   5. 1024x1024

2. Upscale

   1. 4x

   2. Nearest

3. SDXL Refiner 0.9 checkpoint

   1. No offset LoRa

   2. DPM++ 2M SDE Karras

   3. 1024x1024

   4. 0.05-0.08 or 0.2-0.3 denoising strength depending on how much you want it to change

4. Upscale

   1. Same as 2