Furry E621 SFW 7M HQ: 6.92 Million SFW Image Captions for AI Training

CaptionEmporium provides 6.92 million captions for safe-for-work images from the e621/e926 platform, extending to January 2023. The dataset includes captions generated by a large language model (mistralai/Mistral-7B-v0.1) and a multimodal model (THUDM/CogVLM), with 8 LLM and 1 CogVLM caption per image. Most captions are described as substantially larger than 77 tokens.

Use Cases

• Training image captioning models based on the large volume of descriptive text.
• Fine-tuning text-to-image generative models using the described SFW furry art style.
• Benchmarking the quality of AI-generated captions against the provided LLM and CogVLM outputs.
• Studying the characteristics of long-form captions that exceed typical token limits.

Strengths

• Large scale with 6.92 million captions.
• Includes multiple AI-generated caption perspectives (8 LLM + 1 CogVLM per image).
• Captions are noted to be substantially larger than 77 tokens, providing detailed descriptions.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• The dataset is derived from a specific niche community (e621/e926), which may reflect source bias.

Provenance

Source

e621/e926 platform (safe-for-work split).

Collection Method

Captions created by LLMs (mistralai/Mistral-7B-v0.1) and a custom multilabel classifier along with CogVLM (THUDM/CogVLM).

Time Range

Extends to January 2023.

Freshness

Last updated 2024-03-21 01:26:16; freshness should be verified.