The model DOES NOT claim to be "well trained". The dataset was collected and tagged without proper attention to detail.

I recommend always using the tags @dlordstyle, realistic. If the realistic tag makes your generation look like SD 1.5 garbage — that's an issue with the sh1tty merge you're using, not this LoRA.

Due to the poorly tagged dataset, it may be difficult to remove the watermark from generations.

Recommended weight — 0.7.

The model will be removed upon the original artist's request.

All of the images shown are pure generations - no upscaling, inpainting, or other fixes.