GLIDE: Towards Photorealistic Image Generation and Editing with Text-Guided Diffusion Models

ABSTRACT

Diffusion models have recently been shown to generate high-quality synthetic images, especially when paired with a guiding technique to trade off diversity for fidelity. We explore diffusion models for the problem of text-conditional image synthesis and compare two different guidance strategies: CLIP guidance and classifier-free guidance. We find that classifier-free guidance yields higher-quality images that better capture the text prompts. We present GLIDE, a 3.5 billion parameter text-guided diffusion model. Human evaluators overwhelmingly prefer GLIDE over DALL-E 1. Furthermore, we demonstrate that GLIDE can be fine-tuned to perform powerful, zero-shot image editing (inpainting) via natural language prompts.