Multimodal Face Generation Data With Spatial And Semantic Conditioning

MMFace-DiT Dataset provides multimodal conditioning data for high-fidelity, controllable face synthesis. The dataset, created by BharathK333, includes spatial elements like masks and sketches paired with VLM-enriched semantic captions. It was accepted to CVPR 2026 and last updated in April 2026.

Use Cases

• Train diffusion models for face synthesis using spatial masks as conditioning inputs.
• Benchmark text-to-face generation models using the provided VLM-enriched semantic captions.
• Develop controllable face editing pipelines by leveraging the paired sketches and masks.
• Evaluate the impact of multimodal conditioning on output fidelity in face generation tasks.

Strengths

• Dataset components are organized for plug-and-play use with the official MMFace-DiT repository.
• Provides both spatial (masks, sketches) and semantic (captions) conditioning pairs for comprehensive control.

Limitations

• Specific quantitative details like row count, file size, and sample counts are unknown.
• Geographic or demographic coverage of the face data is unspecified, which may introduce bias.

Provenance

Source

Hugging Face dataset repository by BharathK333.

Collection Method

Contains multimodal conditioning data and high-quality captions created for the MMFace-DiT project.

Freshness

Last updated on 2026-04-07.