Anime Caption Danbooru 2021: 5.71 Million SFW Image Captions

CaptionEmporium's anime-caption-danbooru-2021-sfw-5m-hq dataset contains 5.71 million captions for 1.43 million safe-for-work anime-style images from the Danbooru 2021 dataset. Each image has four captions generated by different AI models, including CogVLM and variants of LLaVA-v1.6-34b. The dataset was last updated on Hugging Face in June 2024.

Use Cases

• Training image-to-text models based on the 5.71 million AI-generated captions.
• Comparing caption quality and style across different AI models (CogVLM, LLaVA) for the same image.
• Fine-tuning text encoders or diffusion models for anime content using the described captions.
• Studying the characteristics of AI-generated captions that are substantially larger than 77 tokens.

Strengths

• Large scale with 5.71 million captions covering 1.43 million unique images.
• Provides multiple AI-generated perspectives per image with four distinct caption types.
• Focuses on a curated safe-for-work (SFW) subset of a well-known anime image source.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Last updated 2024-06-09 22:28:57; freshness should be verified.
• The description notes most captions are unsuitable for discrimination using current unspecified methods, which may limit certain applications.

Provenance

Source

CaptionEmporium on Hugging Face, derived from a filtered subset of the Danbooru 2021 dataset.

Collection Method

Captions were generated by AI models (CogVLM, LLaVA-v1.6-34b) applied to the source images.

Freshness

Last updated 2024-06-09 22:28:57.