Uni-Edit: Training Data for Unified Multimodal Model Tuning

Uni-Edit-148k is a collection of 148,000 data points for training Unified Multimodal Models (UMMs) in intelligent image editing. It was created by the Uni-Edit project to serve as a general task for UMM tuning, aiming to overcome task conflicts in conventional multi-task training. The dataset was last updated on May 21, 2026.

Use Cases

• Fine-tuning multimodal models for unified image editing based on the Uni-Edit task paradigm.
• Training models to perform intelligent image manipulation based on the described mutual reinforcement approach.
• Benchmarking generalist multimodal models against a dedicated image editing task.
• Developing end-to-end pipelines for complex image generation and editing workflows.

Strengths

• Contains 148,000 training examples for model tuning.
• Designed specifically for the Uni-Edit task to mitigate inherent task conflicts in multi-task training.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown beyond the 148k figure, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

Uni-Edit project, hosted on Hugging Face.

Collection Method

Likely contains image-text pairs or multimodal prompts for editing tasks, as inferred from the description of the Uni-Edit task.

Freshness

Last updated 2026-05-21 02:06:10; freshness should be verified.