Group Diffusion: Enhancing Image Generation by Unlocking Cross-Sample Collaboration
This addresses a fundamental limitation in diffusion model inference for image generation, offering a novel approach to enhance quality through cross-sample collaboration.
The paper tackles the problem of independent image generation in diffusion models by enabling collaborative generation through shared attention across multiple images during inference, achieving up to 32.2% FID improvement on ImageNet-256x256.
In this work, we explore an untapped signal in diffusion model inference. While all previous methods generate images independently at inference, we instead ask if samples can be generated collaboratively. We propose Group Diffusion, unlocking the attention mechanism to be shared across images, rather than limited to just the patches within an image. This enables images to be jointly denoised at inference time, learning both intra and inter-image correspondence. We observe a clear scaling effect - larger group sizes yield stronger cross-sample attention and better generation quality. Furthermore, we introduce a qualitative measure to capture this behavior and show that its strength closely correlates with FID. Built on standard diffusion transformers, our GroupDiff achieves up to 32.2% FID improvement on ImageNet-256x256. Our work reveals cross-sample inference as an effective, previously unexplored mechanism for generative modeling.