Synthetic Image Collection with Over One Million DALL-E 3 Captions

Over one million AI-generated images paired with high-quality text captions, primarily from DALL-E 3 with contributions from Stable Diffusion and Midjourney v5+. Created by ProGamerGov, the dataset was last updated in October 2024.

Use Cases

• Training text-to-image generation models using the high-quality captions as prompts.
• Fine-tuning image captioning models on the paired image-text data.
• Benchmarking the performance of vision-language models on synthetic content from multiple AI systems.
• Analyzing stylistic differences between images generated by DALL-E 3, Stable Diffusion, and Midjourney.

Strengths

• Contains over one million image-caption pairs.
• Includes content from multiple high-quality AI systems like DALL-E 3 and Midjourney v5+.

Limitations

• Specific image counts per AI system and sample quality metrics are not provided.
• Dataset composition and potential biases from web-scraped sources are unknown.

Provenance

Source

AI-generated images sourced from various websites and individual contributors.

Collection Method

Collected from the web, focusing on DALL-E 3 content and other specified AI systems.

Freshness

Last updated in October 2024.