This is a SDXL based fine-tuned merge model.

VAE is included in the model (VAE selection NONE or Auto Auto).

Please don't use SD1.5 VAE (it will generate noise image).

If you find that the model does not switch, please update webui to version 1.6.

No need to fix face for half body and close-up

This model has special training for traditional costumes.

It is also very good at depicting details and colours.

Thanks to SDXL's superior understanding of natural language.

It is highly recommended that you use natural language to write your cues.

And please forget masterpiece and best quality.

You can even leave out the negative words.

Suggested samplers: DPM++ 2M Karras (I use it most often), Restart (my fingers don't crumble as easily).

Step count suggestion: 20+

CFG recommendation: 3-11

Size suggestion (official SDXL suggestion):

1024×1024 (1:1)

Vertical version

640×1536 (9:21)

768×1344 (9:16)

832×1216 (2:3)

896×1152 (4:5)

For full body images or images with a small head, it is recommended to use the ADetailer plugin to fix the face.