AdaFace: A Versatile Text-space Face Encoder for Face Synthesis and Processing

Official demo for our working paper AdaFace: A Versatile Text-space Face Encoder for Face Synthesis and Processing.

❗️What's New❗️

  • Support switching between three model styles: Photorealistic, Realistic and Anime.
  • If you just changed the model style, the first image/video generation will take extra 20~30 seconds for loading the new model weight.

❗️Tips❗️

  1. Upload one or more images of a person. If multiple faces are detected, we use the largest one.
  2. "Highlight face" will make the face more prominent in the generated image.
  3. There are 3 "Composition Levels". Increasing the level will improve the composition at the cost of the face quality. Some difficult prompts like "In a wheelchair" and "on a horse" need at least level 1.
  4. AdaFace Text-to-Video: AdaFace-Animate
Gender

Gender prefix. Select only when the model errs.
Prompt

Try something like 'walking on the beach'. If the face is not in focus, try checking 'Highlight face'.

Enhance the facial features by prepending 'face portrait' to the prompt
0 2
1 8
Base Model Style Type

Switching the base model type will take 10~20 seconds to reload the model
1 8
0 2147483647

Uncheck for reproducible results