Imagen

Imagen is an AI system created by Google Research in 2022 for controllable image generation based solely on text prompts. Key features:

• Uses a deep diffusion model for synthesis.
• Provides precise control over generated images.
• Avoids inappropriate or harmful content.
• High-resolution outputs with photorealism.
• Conditioned on text captions during training.
• Builds on aligned image-text datasets.
• Unparalleled textual understanding.
• Can render detailed avatars, landscapes, artwork.
• Control extends across artistic styles, poses, lighting.

Imagen showcases major improvements in textual understanding and fine-grained controllability compared to prior text-to-image models. It also employs safety measures during training and sampling.

See also:

• Text-to-image synthesis
• Diffusion model
• Stable Diffusion
• DALL-E