NEW: In version v2, I expanded the Lora dataset and fixed some issues. Now the backgrounds are more detailed, the colors have improved, and the overall style looks much more interesting. Although the style has changed, the old version still feels unique and valuable to me. Testing showed excellent results with a weight of 1.0. In the sample images, I didn’t use an upscaler, high-res fix, or quality tags so you could see the style itself.

Note: I labeled version V2 as SDXL instead of IllustriousXL so that you can, if desired, use it in an online generator together with NoobAI.

-----------------------------------------------------------------------------

Since I couldn’t find a LoRA in the style I wanted, I decided to create my own. It’s a rather cute style of non-professional drawing with sharp outlines.

During testing, I found that the best weight is between 0.7 and 1, though you can try 0.4 if you need a more detailed image.

In general, I created what I wanted and just decided to share it publicly :)

-----------------------------------------------------------------------------

Since the original dataset had slightly imperfect hands, this might result in crooked hands in the generated image. Therefore, add the tag "bad hands, crooked fingers" to the negative prompts.

I uploaded 4 versions: v2, ep11 - 11th epoch, ep14 - 14th epoch, and ep15 - 15th epoch. Test each one and find the version that works best for you.

P.S: I generated the sample images without quality tags, high-res fixes, or upscales to showcase the original style.

描述:

训练词语: