RealGen: Photorealistic Text-to-Image Generation via Detector-Guided Rewards
The model presented in the paper RealGen: Photorealistic Text-to-Image Generation via Detector-Guided Rewards proposes a photorealistic text-to-image framework. RealGen integrates an LLM component for prompt optimization and a diffusion model for realistic image generation. It introduces a "Detector Reward" mechanism, which quantifies artifacts and assesses realism using both semantic-level and feature-level synthetic image detectors. This reward signal is leveraged with the GRPO algorithm to optimize the entire generation pipeline, significantly enhancing image realism and detail.
Project Page: https://yejy53.github.io/RealGen/ Code: https://github.com/yejy53/RealGen
Detection Models
RealGen utilizes specialized detection models to guide its generation process:
- Semantic Detector: Forensic-Chat, a generalizable and interpretable detector optimized from Qwen2.5-VL-7B. It assesses authenticity by analyzing image content (e.g., smooth greasy skin, artifacts in faces/hands, unnatural background blur).
- Feature Detector: OmniAID achieves stable and accurate detection by being pre-trained on large-scale real and synthetic datasets. Feature-level artifacts are primarily associated with frequency artifacts and abnormal noise patterns.
Citation
If you find our work helpful or inspiring, please feel free to cite it.
@article{ye2025realgen,
title={RealGen: Photorealistic Text-to-Image Generation via Detector-Guided Rewards},
author={Ye, Junyan and Zhu, Leqi and Guo, Yuncheng and Jiang, Dongzhi and Huang, Zilong and Zhang, Yifan and Yan, Zhiyuan and Fu, Haohuan and He, Conghui and Li, Weijia},
journal={arXiv preprint arXiv:2512.00473},
year={2025}
}