Abstract
Diffusion Language Models (DLMs) have achieved substantial progress in natural language generation tasks. Recent research shows that adaptive token generation ordering can significantly improve performance in mathematical reasoning and code synthesis applications. This work investigates the optimization of generation order for both text-to-image synthesis and multimodal understanding.
We first establish that, unlike structured problems in language generation such as Sudoku puzzles, model logits alone are insufficient to determine optimal generation sequences in text-to-image generation and multimodal understanding. To address this challenge, we introduce a learnable control module trained via Group Relative Policy Optimization (GRPO) to determine the generation order.
Our results demonstrate that learning this control block substantially improves both text-to-image alignment and multimodal understanding in DLMs. In particular, it enhances the model's ability to capture fine-grained spatial relationships in generated images while also strengthening performance on multimodal reasoning and comprehension tasks.
We evaluate our framework on GenEval, an object-focused benchmark for text-to-image alignment, where it achieves 4.08% relative improvements. Additionally, experiments on VLMEvalKit confirm 4.85% relative improvements in multimodal understanding, highlighting the broad effectiveness of our approach.
Blogger's Review: This paper successfully addresses the order optimization problem in multimodal generation by introducing a learnable control module, showcasing potential in text-to-image and multimodal understanding tasks. It presents new insights for future research and deserves attention.