Fine-T2I: Fine-Tuning Text-to-Image Dataset

Fine-T2I is a text-to-image dataset containing between 100,000 and 1,000,000 image-caption pairs, released in February 2026 by researchers at Northeastern University. It provides high-quality visual and textual data specifically curated for fine-tuning generative models and is distributed in WebDataset format.

Use Cases

• Fine-tuning text-to-image models using the image and caption pairs
• Training image-to-text captioning systems using the text modality
• Evaluating generative model alignment between the image and image-generation fields

Strengths

• 100K to 1M record count
• Apache 2.0 license for open redistribution
• WebDataset format support for efficient large-scale data streaming

Limitations

• Lack of explicit column schema documentation in the primary metadata
• Potential for high computational and storage overhead due to the image-heavy modality

Provenance

Source

Northeastern University (Xu Ma, Yitian Zhang, Qihua Dong, Yun Fu)

Freshness

Last updated February 20, 2026.

Geography

United States