Search papers, labs, and topics across Lattice.
This paper introduces Flex-Forcing, a novel framework that integrates autoregressive and bidirectional generation methods for video diffusion models, addressing the limitations of existing approaches in inference speed and long-range consistency. By employing a flexible chunking mechanism that adapts to device budgets and allows for bidirectional inference across chunks, the model enhances global coherence while maintaining efficient frame synthesis. Extensive evaluations reveal that Flex-Forcing significantly outperforms traditional models in video quality and stability, while also accelerating inference times.
Flex-Forcing achieves superior video generation quality and stability by unifying autoregressive and bidirectional methods, all while speeding up inference.
Recent progress in large-scale generative models has substantially advanced video generation, yet existing methods remain constrained by a rigid inference paradigm. Bidirectional diffusion models excel at global coherence and visual fidelity but suffer from slow inference, while autoregressive models offer efficient and streaming generation at the cost of long-range consistency and exposure bias. We introduce Flex-Forcing, a unified training and inference framework that enables a video diffusion model to seamlessly operate under both bidirectional and autoregressive generation regimes. The core idea is a flexible chunking mechanism jointly defined over the temporal axis and denoising steps. This design allows the model to (1) perform flexible chunking according to different device budgets, (2) perform bidirectional inference across chunks for global structure planning, while generating frames autoregressively within each chunk for efficient and fine-grained synthesis, and (3) perform any-order, any-timestep autoregressive generation without the strict causal constraint. Extensive experiments on multiple video generation benchmarks demonstrate that Flex-Forcing achieves consistently better video quality, long-video stability than strong baselines with a rigid inference schedule, while offering faster inference.
