MegaStyle-1.4M: 1.4 Million Images for Fine-Grained Style Consistency

MegaStyle-1.4M is a large-scale dataset of 1.4 million synthetic images designed for style consistency. It was generated by Tencent using a pipeline that maps 170,000 style prompts and 400,000 content prompts via the Qwen-Image model. The dataset was last updated on the Hugging Face platform in April 2026.

Use Cases

• Training text-to-image models for style consistency based on the described prompt mapping pipeline.
• Benchmarking intra-style consistency in generated images based on the dataset's stated design goal.
• Researching fine-grained style representation and disentanglement based on the diverse style prompts mentioned.
• Developing conditional image generation systems using the combination of style and content prompts described.

Strengths

• Contains 1.4 million high-quality generated images, providing a large-scale resource.
• Built from 170,000 curated style prompts, suggesting diversity in fine-grained styles.
• Combines style and content prompts, with 400,000 content prompts for varied subject matter.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data may reflect bias inherent to the generative model and prompt sources used.

Provenance

Source

Tencent, via Hugging Face.

Collection Method

Generated through a scalable pipeline leveraging consistent text-to-image style mapping of Qwen-Image.

Freshness

Last updated 2026-04-20 09:03:50; freshness should be verified.