Read It Back: Pretrained MLLMs Are Zero-Shot Reward Models for Text-to-Image Generation

2026-07-13

AI Summary

This work addresses the challenge of constructing reinforcement learning reward signals for text-to-image generation without requiring training or human preference labels. It proposes SpectraReward, which leverages a pretrained multimodal large language model (MLLM) as a zero-shot reward model, measuring text-image alignment directly via the average log-likelihood of the original prompt obtained through a single teacher-forced forward pass conditioned on the generated image. Furthermore, it introduces Self-SpectraReward, which establishes a closed-loop self-optimization mechanism within a unified model by coupling generative and understanding branches, thereby achieving intrinsic alignment between policy and reward. Evaluated across five out-of-distribution benchmarks, the method significantly improves generation quality, with Self-SpectraReward matching or even surpassing the performance of substantially larger external reward models, demonstrating its effectiveness and generalizability.

Abstract

In this paper, we propose SpectraReward, a training-free reward function that turns pretrained MLLMs into off-the-shelf reward models for image-generation reinforcement learning. Instead of asking the MLLM to judge a generated image or answer decomposed verification questions, SpectraReward measures how well the original prompt can be recovered from the generated image through a single image-conditioned, teacher-forced forward pass. We use the average image-conditioned prompt log-likelihood as the reward, directly reusing the MLLM's pretrained image-text alignment ability without preference labels, reward-model fine-tuning. We further introduce Self-SpectraReward, a special case for unified multimodal models where the policy's own understanding branch serves as the reward model for its generation branch, forming a closed-loop self-improving framework without external reward models or external knowledge. Extensive experiments validate SpectraReward through a broad image-generation RL study covering two diffusion models, three RL algorithms, nine reward MLLM backbones from four MLLM families spanning 4B to 235B parameters, and five out-of-distribution text-to-image benchmarks. Results show that both SpectraReward and Self-SpectraReward significantly and consistently improve generation performance and outperform prior MLLM-derived reward training methods. Further analysis reveals that larger reward MLLMs are not always better, while Self-SpectraReward can match or surpass much larger external reward models, suggesting that reward-policy alignment is a key factor for effective image-generation RL. Project Page: https://huangrh99.github.io/SpectraReward/

Problem

• text-to-image generation
• reward modeling
• multimodal large language models
• reinforcement learning
• zero-shot

Innovation

• SpectraReward
• zero-shot reward model
• text-to-image generation
• reinforcement learning
• self-improving framework