Unlocking Complex Visual Generation via Closed-Loop Verified Reasoning

TL;DR

The paper introduces CLVR, a novel framework that combines visual-language reasoning with pixel-level diffusion to improve complex text-to-image generation, addressing current limitations in planning, verification, and latency.

Contribution

CLVR offers a comprehensive system integrating visual verification, reinforcement learning, and weight merging to enhance reasoning, stability, and efficiency in complex visual generation tasks.

Findings

CLVR outperforms existing open-source baselines on multiple benchmarks.

It approaches the performance of proprietary commercial models.

The framework enables scalable, test-time complex visual generation.

Abstract

Despite rapid advancements, current text-to-image (T2I) models predominantly rely on a single-step generation paradigm, which struggles with complex semantics and faces diminishing returns from parameter scaling. While recent multi-step reasoning approaches show promise, they are hindered by ungrounded planning hallucinations lacking verification, monolithic post-hoc reflection, long-context optimization instabilities, and prohibitive inference latency. To overcome these bottlenecks, we propose the Closed-Loop Visual Reasoning (CLVR) framework, a comprehensive system that deeply couples visual-language logical planning with pixel-level diffusion generation. CLVR introduces an automated data engine with step-level visual verification to synthesize reliable reasoning trajectories, and proposes Proxy Prompt Reinforcement Learning (PPRL) to resolve long-context optimization instabilities by distilling interleaved multimodal histories into explicit reward signals for accurate causal attribution. Furthermore, to mitigate the severe latency bottleneck caused by iterative denoising, we propose $\Delta$-Space Weight Merge (DSWM), a theoretically grounded method that fuses alignment weights with off-the-shelf distillation priors, reducing the per-step inference cost to just 4 NFEs without requiring expensive re-distillation. Extensive experiments demonstrate that CLVR outperforms existing open-source baselines across multiple benchmarks and approaches the performance of proprietary commercial models, unlocking general test-time scaling capabilities for complex visual generation.