Causal Forcing++: Scalable Few-Step Autoregressive Diffusion Distillation for Real-Time Interactive Video Generation

Hugging Face Daily Papers Papers
few-step autoregressive diffusion-distillation video-generation real-time causal-consistency world-model

Summary

Causal Forcing++ presents a novel causal consistency distillation method for frame-wise autoregressive video generation, achieving state-of-the-art quality with reduced latency and training cost.

Real-time interactive video generation requires low-latency, streaming, and controllable rollout. Existing autoregressive (AR) diffusion distillation methods have achieved strong results in the chunk-wise 4-step regime by distilling bidirectional base models into few-step AR students, but they remain limited by coarse response granularity and non-negligible sampling latency. In this paper, we study a more aggressive setting: frame-wise autoregression with only 1--2 sampling steps. In this regime, we identify the initialization of a few-step AR student as the key bottleneck: existing strategies are either target-misaligned, incapable of few-step generation, or too costly to scale. We propose Causal Forcing++, a principled and scalable pipeline that uses causal consistency distillation (causal CD) for few-step AR initialization. The core idea is that causal CD learns the same AR-conditional flow map as causal ODE distillation, but obtains supervision from a single online teacher ODE step between adjacent timesteps, avoiding the need to precompute and store full PF-ODE trajectories. This makes the initialization both more efficient and easier to optimize. The resulting pipeline, \ours, surpasses the SOTA 4-step chunk-wise Causal Forcing under the \textbf{frame-wise 2-step setting} by 0.1 in VBench Total, 0.3 in VBench Quality, and 0.335 in VisionReward, while reducing first-frame latency by 50\% and Stage 2 training cost by sim4times. We further extend the pipeline to action-conditioned world model generation in the spirit of Genie3. Project Page: https://github.com/thu-ml/Causal-Forcing and https://github.com/shengshu-ai/minWM .
Original Article
View Cached Full Text

Cached at: 05/15/26, 04:23 AM

Paper page - Causal Forcing++: Scalable Few-Step Autoregressive Diffusion Distillation for Real-Time Interactive Video Generation

Source: https://huggingface.co/papers/2605.15141

Abstract

A novel causal consistency distillation method enables efficient frame-wise video generation with reduced latency and improved quality compared to existing chunk-wise approaches.

Real-time interactive video generation requires low-latency, streaming, and controllable rollout. Existing autoregressive (AR)diffusion distillationmethods have achieved strong results in the chunk-wise 4-step regime by distilling bidirectional base models into few-step AR students, but they remain limited by coarse response granularity and non-negligible sampling latency. In this paper, we study a more aggressive setting:frame-wise autoregressionwith only 1--2 sampling steps. In this regime, we identify the initialization of a few-step AR student as the key bottleneck: existing strategies are either target-misaligned, incapable of few-step generation, or too costly to scale. We propose Causal Forcing++, a principled and scalable pipeline that usescausal consistency distillation(causal CD) forfew-step AR initialization. The core idea is thatcausal CDlearns the same AR-conditional flow map as causal ODE distillation, but obtains supervision from a single online teacher ODE step between adjacent timesteps, avoiding the need to precompute and store full PF-ODE trajectories. This makes the initialization both more efficient and easier to optimize. The resulting pipeline, \ours, surpasses the SOTA 4-step chunk-wise Causal Forcing under the \textbf{frame-wise 2-step setting} by 0.1 inVBenchTotal, 0.3 inVBenchQuality, and 0.335 inVisionReward, while reducing first-frame latency by 50\% and Stage 2 training cost by sim4times. We further extend the pipeline to action-conditionedworld model generationin the spirit ofGenie3. Project Page: https://github.com/thu-ml/Causal-Forcing and https://github.com/shengshu-ai/minWM .

View arXiv pageView PDFProject pageAdd to collection

Get this paper in your agent:

hf papers read 2605\.15141

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2605.15141 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2605.15141 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2605.15141 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

Lip Forcing: Few-Step Autoregressive Diffusion for Real-time Lip Synchronization

Hugging Face Daily Papers

This paper introduces Lip Forcing, the first autoregressive diffusion method for real-time video-to-video lip synchronization. By distilling a 14B teacher into causal students and using only two denoising steps, it achieves 31 FPS streaming at 1.3B scale, 17.6x faster than same-scale bidirectional models.