Scalable Diffusion Models with Transformers
This addresses the scalability challenge in diffusion models for high-quality image generation, though it is incremental as it adapts an existing transformer approach to a known bottleneck.
The paper tackles the problem of scaling diffusion models for image generation by replacing the U-Net backbone with a transformer architecture, resulting in state-of-the-art performance with an FID of 2.27 on the class-conditional ImageNet 256x256 benchmark.
We explore a new class of diffusion models based on the transformer architecture. We train latent diffusion models of images, replacing the commonly-used U-Net backbone with a transformer that operates on latent patches. We analyze the scalability of our Diffusion Transformers (DiTs) through the lens of forward pass complexity as measured by Gflops. We find that DiTs with higher Gflops -- through increased transformer depth/width or increased number of input tokens -- consistently have lower FID. In addition to possessing good scalability properties, our largest DiT-XL/2 models outperform all prior diffusion models on the class-conditional ImageNet 512x512 and 256x256 benchmarks, achieving a state-of-the-art FID of 2.27 on the latter.