DeepFloyd IF

lastmod 2023-04-30created 2023-04-302 edits

Open-source Text-to-Image model

Features

• New image model from Stability AI
• Higher language understanding than existing models, and generates photorealistic images
• Combines a frozen text encoder based on the T5 Transformer with a three-stage pixel diffusion module
  • 64x64, 256x256, 1024x1024px
• Zero-shot FID score of 6.66 (COCO dataset)

See also

• Text-to-Image
• Stable Diffusion

Favorite site

• Github - deep-floyd/IF