Ask, Solve, Generate: Self-Evolving Unified Multimodal Understanding and Generation via Self-Consistency Rewards

2026-06-25

AI Summary

This work addresses the limitation of existing unified multimodal large models, which rely on human annotations or external rewards for visual understanding and generation tasks and lack mechanisms for autonomous evolution using only unlabeled images. The authors propose a self-evolution training framework that coordinates three roles—questioner, answerer, and generator—and trains solely on self-consistency signals without human supervision. A key innovation is the introduction of Solver Token Entropy (STE) as a continuous difficulty metric, combined with multi-scale question-answering fidelity and cycle-consistent caption evaluation to jointly optimize comprehension and generation capabilities. The method is adaptable to BLIP3o, BAGEL, and VARGPT-v1.1 architectures, consistently outperforming baselines across eight understanding benchmarks—e.g., improving BAGEL by 3.5% on MMMU—and boosting GenEval image generation performance from 82% to 85%.

Abstract

Most unified large multimodal models (LMMs) that support both visual understanding and image generation still rely on curated post-training supervision, such as human annotations, preference labels, or external reward models. We ask whether a unified LMM can improve both abilities autonomously using only unlabeled images. We propose a self-evolving training framework with three internal roles: a Proposer that generates visual questions, a Solver that answers and evaluates them, and a Generator that synthesizes images. Training uses only self-derived consistency signals, without human annotations, preference labels, or task-trained external reward/judge models. To stabilize learning, we introduce Solver Token Entropy (STE), a continuous difficulty signal based on token-level prediction uncertainty that remains useful even when sample-level consistency becomes unreliable. For image generation, we design a multi-scale internal evaluation scheme that combines question-answer fidelity scoring with cycle-consistent captioning. This creates a solver-mediated coupling, where better visual understanding enables more reliable generation assessment and stronger internal training signals. The framework preserves the same role decomposition, reward logic, and training schedule across diffusion-based BLIP3o, rectified-flow BAGEL, and autoregressive VARGPT-v1.1 architectures, requiring only each backbone's native prompting and generation interface. Across eight understanding metrics, our method consistently improves over the corresponding base models. On BAGEL, it achieves a $+3.5\%$ absolute gain on MMMU and improves GenEval image generation performance from $82\%$ to $85\%$. Code and models are publicly released.

Problem

• unified multimodal models
• self-evolving
• visual understanding
• image generation
• self-consistency rewards

Innovation

• self-evolving
• self-consistency rewards
• unified multimodal model
• Solver Token Entropy
• multi-scale internal evaluation