Diffusion Forcing: Next-token Prediction Meets Full-Sequence Diffusion

By Boyuan Chen, Diego Marti Monso, Yilun Du, Max Simchowitz, Russ Tedrake, Vincent Sitzmann. Neural Information Processing Systems, 2024

584 citations recorded by the source corpus.

This paper presents Diffusion Forcing, a new training paradigm where a diffusion model is trained to denoise a set of tokens with independent per-token noise levels. We apply Diffusion Forcing to sequence generative modeling by training a causal next-token prediction model to generate one or several future tokens without fully diffusing past ones. Our approach is shown to combine the strengths of next-token prediction models, such as variable-length generation, with the strengths of full-sequence diffusion models, such as the ability to guide sampling to desirable trajectories. Our method offers a range of additional capabilities, such as (1) rolling-out sequences of continuous tokens, such as video, with lengths past the training horizon, where baselines diverge and (2) new sampling and guiding schemes that uniquely profit from Diffusion Forcing's variable-horizon and causal architecture, and which lead to marked performance gains in decision-making and planning tasks. In addition to its empirical success, our method is proven to optimize a variational lower bound on the likelihoods of all subsequences of tokens drawn from the true joint distribution. Project website: https://boyuan.space/diffusion-forcing

Subfields: Genetics, Materials Chemistry, Nuclear and High Energy Physics

View Paper: https://www.semanticscholar.org/paper/40d63dc2b465c9081e4efc5a19514da151e97fe7

Citations in the corpus

• Self Forcing: Bridging the Train-Test Gap in Autoregressive Video Diffusion (2025, unknown): “While some approaches attempt to mitigate this issue in video diffusion models by incorporating noisy context frames during inference [8, 11, 105], such design sacrifices temporal consistency, complicates the KV-cache design, increases generation latency, and does not fundamentally resolve the…”
• Pyramidal Flow Matching for Efficient Video Generative Modeling (2024, unknown): “An alternative line of research integrates diffusion models with autoregressive modeling (Chen et al., 2024a; Valevski et al., 2024) to natively support long video generation, but is still limited in context length and training efficiency.”
• Emerging Properties in Unified Multimodal Pretraining (2025, unknown): “For interleaved multi-image generation, we adopt the diffusion forcing strategy [8], which adds independent noise levels to different images and conditions each image on noisy representations of preceding images.”
• From Slow Bidirectional to Fast Autoregressive Video Diffusion Models (2024, unknown): “Instead of naively distilling an autoregressive diffusion model [8, 28] into a few-step student, we propose an asymmetric distillation strategy where we distill the knowledge in a pretrained teacher diffusion model with bidirectional attention into our causal student model.”