In txt2img, you will only get low quality results due to the training data's quality being bad.

Please use img2img without the LoRA to fix the quality issue.

You can't get the same quality in just one generation.

My workflow:

1. txt2img for base image.

2. img2img + 2x upscale without the lora to fix glitches.

3. img2img + 2x upscale with the lora to enhance details.