SD3-univeral beta β

Continuing training from beta alpha

SD3-univeral beta α

• The model fine-tuning training was completed on SD3medium-2B, using a dataset of approximately 50,000 real photographs for multiple rounds of full

• optimized the oversaturation issue and strange stripe problem that existed in SD3-2B ,

• extremely realistic images with rich levels of

• can generate photos with high color dynamic range and excellent light-shadow relationships.

For prompts, please use natural language descriptions. It is recommended to use the prompt expansion feature in the tensor workbench, as simple tags cannot generate high-quality images.