The V8 version differs from previous versions. The base model has been retrained using the latest GPT-4 low-quality captions and Cog natural language completion for details. Meanwhile, hand fine-tuning, lighting perception, and color rendering have also been retrained and integrated into this retrained base model.

Recommended Sampling Steps: 25–30 steps

Recommended Sampler: DPM++ 2M Karras

Name in local WebUI 1.9.0: DPM++ 2M

Recommended Upscaler: R-ESRGAN_4x+ Anime6B

Enable After Detailer when needed: face_yolov8n_v2.pt

Reposted from: Wanxiang XL Advanced_V80 (Official Release)