MultiRef: A Large-Scale Dataset for Multi-Reference Image Generation

38,076 synthetically generated samples for training controllable image generation models to use multiple visual references. The dataset was created by ONE-Lab to facilitate research in generative AI and was last updated on November 26, 2025.

Use Cases

• Training controllable image generation models based on the dataset's multiple visual references.
• Benchmarking model performance on multi-reference synthesis tasks based on the dataset's scale.
• Developing methods for visual element composition based on the dataset's synthetically generated samples.

Strengths

• Large scale with over 38,000 samples.
• Specifically designed for the multi-reference image generation task.
• Synthetically generated, which may allow for controlled variation.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• The description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

ONE-Lab via Hugging Face.

Collection Method

All samples are synthetically generated.

Time Range

null

Freshness

Last updated 2025-11-26 09:41:39; freshness should be verified.

Geography

null