XXMix_9 Realism SDX

Version v1

Trigger Word: xxmixgirl-------It's okay without trigger words, but it's better to include them.

Reverse Prompts: (worst quality, low quality, illustration, 3d, 2d, painting, cartoons, sketch), tooth, open mouth, bad hand, bad fingers,

Image Size: 768*1280

Full Body: It is recommended to open in high resolution and use ad to modify the face.

Portrait: It is not necessary to open in high resolution, you can use ad to modify the face

Sampling Times: 30

CFG Scale: 8-10

• -----------------------------------------------------------------------------------------------------------------------------
• Currently, the character distortion is strong, and the skin texture details are not clear enough. I hope to improve these issues through continuous iterations.
• The reverse prompts that currently work best for this XL model are: (worst quality, low quality, illustration, 3d, 2d, painting, cartoons, sketch), tooth, open mouth,
• XXMix_9realisticSDXL is a fine-tuning model based on the Stable Diffusion XL model, aiming to improve the poor performance of Stable Diffusion XL in the attractiveness of Asian female characters. The current v1 version is still in the experimental stage with many issues, but I will continue to iterate and optimize.
• Trigger Words: xxmix girl woman (in some cases, it may be necessary to increase the weight of keywords)

Known issues at present include:

1. There is a chance that the character's name or art style may not take effect on the face shape. In this case, try increasing the weight of the trigger words.

2. The model is trained on a canvas of 1024*1024, so images with different proportions may result in higher error rates. It is recommended to use a canvas of 1024*1024 as much as possible. I will try training with datasets of different canvases in future iterations.

3. Regarding the reverse prompts: The reverse prompts we currently use help maintain the model's good performance in Asian facial effects. If you try using different reverse prompts, the effect may not be as good as the prompts I currently use. However, if you find reverse prompts that are more suitable for this large model, I hope you can share them with everyone.

4. Due to the aspect ratio, this model is currently more suitable for generating half-length or portrait images. If full-body images are needed, it is recommended to enable the high-definition restoration function or use the faterdetail extension plugin.

Trigger Word: xxmixgirl-------It's okay without trigger words, but it's better to include them.

Reverse Prompts: (worst quality, low quality, illustration, 3d, 2d, painting, cartoons, sketch), tooth, open mouth, bad hand, bad fingers,

Image Size: 768*1280

Full Body: It is recommended to open in high resolution and use ad to modify the face.

Portrait: It is not necessary to open in high resolution, you can use ad to modify the face

Sampling Times: 30

CFG Scale: 8-10