FreeStyle Dataset: CRef/SRef LoRA Triplets for Diffusion Training

A collection of CRef/SRef LoRA triplets exported from the 0426 diffusion training data. Each training example contains three images: a content reference, a style reference, and a target image generated from the combined condition. The dataset was created by Blue2Giant and last updated on June 17, 2026.

Use Cases

• Training LoRA adapters for diffusion models based on content and style reference images.
• Researching disentanglement of content and style in image generation based on the triplet structure.
• Benchmarking image generation quality from combined content and style conditions.
• Developing or evaluating multi-reference image generation techniques.

Strengths

• Provides a structured triplet format (content, style, target) for each training example.
• Includes deduplicated metadata and provenance lookup files at the image level.
• Dataset was last updated on 2026-06-17, suggesting recent maintenance.

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.
• Row count, file formats, and license information are unknown, which may limit suitability assessment.

Provenance

Source

huggingface

Collection Method

Exported from the 0426 diffusion training data.

Freshness

Last updated 2026-06-17 15:26:33; freshness should be verified.