Switti: Designing Scale-Wise Transformers for Text-to-Image Synthesis

1Yandex Research        2HSE University        3MIPT        4Skoltech        5AIRI        6ITMO
MY ALT TEXT

Abstract

This work presents Switti, a scale-wise transformer for text-to-image generation. We start by adapting an existing next-scale prediction autoregressive (AR) architecture to T2I generation, investigating and mitigating training stability issues in the process. Next, we argue that scale-wise transformers do not require causality and propose a non-causal counterpart facilitating ∼21% faster sampling and lower memory usage while also achieving slightly better generation quality. Furthermore, we reveal that classifier-free guidance at high-resolution scales is often unnecessary and can even degrade performance. By disabling guidance at these scales, we achieve an additional sampling acceleration of ∼32% and improve the generation of fine-grained details. Extensive human preference studies and automated evaluations show that Switti outperforms existing T2I AR models and competes with state-of-the-art T2I diffusion models while being up to 7× faster.

Human evaluation

MY ALT TEXT

Switti vs competing AR and diffusion-based models.



Inference performance evaluation

MY ALT TEXT

Comparison of models' 1024×1024 image generation time.



Automated Metrics

MY ALT TEXT

Quantitative comparison of Switti to competing AR and diffusion-based models.

The best model is highlighted in red, the second-best in blue, and the third-best in yellow according to the respective automated metric.

BibTeX


        @article{voronov2024switti,
          title={Switti: Designing Scale-Wise Transformers for Text-to-Image Synthesis},
          author={Voronov, Anton and Kuznedelev, Denis and Khoroshikh, Mikhail and Khrulkov, Valentin and Baranchuk, Dmitry},
          journal={arXiv preprint arXiv:2412.01819},
          year={2024}
        }