Open Preference Dataset for Text-to-Image Generation on Hugging Face Hub

AI Impact Summary

Hugging Face's Data is Better Together team released an Apache 2.0 licensed open-image-preferences dataset for text-to-image generation. The dataset covers multiple categories and model families, with toxicity filtering, manual review, and synthetic prompt enhancement, and is generated using stabilityai/stable-diffusion-3.5-large and black-forest-labs/FLUX.1-dev via the Diffusers workflow, with results and prompts published on the Hugging Face Hub. This provides open-source teams a concrete benchmark and a fine-tuning resource to improve alignment with human preferences and to experiment with LoRA adapters across model families.