PortraitGen: Exemplar-Driven GRPO with Dual-Reward Guidance for Photorealistic Portrait Generation

2026-06-25

AI Summary

This work addresses the limitations of existing GRPO-based post-training methods for text-to-image generation, which often produce AI artifacts and biologically implausible facial details, hindering high-fidelity portrait synthesis. To overcome this, the study introduces real human portrait images directly into the GRPO sampling process and proposes a dual-reward mechanism tailored for portraiture—integrating OmniReward with a novel AI-Portrait reward. By leveraging image inversion and latent-space sampling within the Group Relative Policy Optimization framework, the approach steers the generative model beyond its original output boundaries. Experiments on the newly curated PortraitBench benchmark demonstrate that the proposed method significantly outperforms current state-of-the-art techniques, achieving unprecedented levels of photorealism and biological plausibility in generated facial details.

Abstract

Reinforcement Learning like Group Relative Policy Optimization (GRPO) has significantly advanced text-to-image post-training. However, current methods often favor superficial aesthetics, such as over-saturated colors, leaving critical flaws like AI artifacts and biological implausibilities unresolved. We attribute these limitations to two primary factors: (1) The absence of real images during post-training confines GRPO sampling to the original distribution, failing to break inherent generative boundaries; (2) the optimization process lacks specific rewards targeting fine-grained artifacts like overly oily skin and other AI artifacts. To address this, we propose PortraitGen, a novel framework tailored for photorealistic portrait generation. First, we break inherent generative boundaries by directly introducing real images into the GRPO sampling groups, where image inversion is employed to obtain their transition probabilities and latents. Second, to explicitly steer the model toward photorealism, we introduce a complementary dual-reward mechanism: OmniReward for general quality and AI-Portrait for human-centric fidelity. Furthermore, we curate PortraitBench, a comprehensive portrait-centric benchmark. Extensive experiments demonstrate that PortraitGen significantly outperforms existing baselines, effectively suppressing AI artifacts and achieving unprecedented photorealism.

Problem

• AI artifacts
• photorealistic portrait generation
• biological implausibilities
• text-to-image post-training
• generative boundaries

Innovation

• GRPO
• exemplar-driven
• dual-reward
• photorealistic portrait generation
• image inversion