Improving Image Generation with Better Captions

Abstract

We study how image generation models can be improved by training on better image captions. We develop an automatic captioning pipeline that generates highly descriptive image captions. Training text-to-image models on these improved captions substantially improves the quality and prompt-following ability of the resulting models, which we call DALL-E 3.

Key Findings

• 1Showed that better training captions dramatically improve image generation quality
• 2Developed an automatic captioning pipeline for training data improvement
• 3Achieved significantly better prompt-following compared to DALL-E 2
• 4Demonstrated improved text rendering in generated images
• 5Integrated natively with ChatGPT for conversational image generation

Impact & Significance