DiT-Air: Revisiting the Efficiency of Diffusion Model Architecture Design in Text to Image Generation
This work addresses efficiency challenges in text-to-image generation for AI practitioners, but it is incremental as it builds on existing DiT architectures.
The authors tackled the problem of improving efficiency in diffusion model architectures for text-to-image generation, finding that a standard DiT variant is comparable to specialized models and more parameter-efficient, with DiT-Air achieving state-of-the-art performance on benchmarks like GenEval and T2I CompBench.
In this work, we empirically study Diffusion Transformers (DiTs) for text-to-image generation, focusing on architectural choices, text-conditioning strategies, and training protocols. We evaluate a range of DiT-based architectures--including PixArt-style and MMDiT variants--and compare them with a standard DiT variant which directly processes concatenated text and noise inputs. Surprisingly, our findings reveal that the performance of standard DiT is comparable with those specialized models, while demonstrating superior parameter-efficiency, especially when scaled up. Leveraging the layer-wise parameter sharing strategy, we achieve a further reduction of 66% in model size compared to an MMDiT architecture, with minimal performance impact. Building on an in-depth analysis of critical components such as text encoders and Variational Auto-Encoders (VAEs), we introduce DiT-Air and DiT-Air-Lite. With supervised and reward fine-tuning, DiT-Air achieves state-of-the-art performance on GenEval and T2I CompBench, while DiT-Air-Lite remains highly competitive, surpassing most existing models despite its compact size.