SpeedVFI: One-step Diffusion for Efficient Video Frame Interpolation

Generative video diffusion models have shown strong robustness to large motion and occlusions for video frame interpolation (VFI). However, their inference efficiency lags significantly behind learning-based methods due to the structural redundancy of pairwise inference and the procedural latency of multi-step iterative denoising. To address these limitations, we propose SpeedVFI, a task-specific one-step diffusion formulation that recasts generative VFI as unified sequence interpolation. SpeedVFI achieves dual efficiency improvements by interpolating the entire video sequence in a single forward pass to eliminate pairwise overhead, and by distilling the generation trajectory into a one-step denoising process to bypass iterative latency. To make this formulation effective for VFI, we introduce temporal RoPE alignment for temporally consistent conditioning and noise-centric partial attention to reduce computational overhead while preserving global context. Extensive experiments demonstrate that SpeedVFI accelerates diffusion-based VFI by orders of magnitude while maintaining competitive quantitative and visual quality.

Many videos need new frames inserted between existing frames, for example to create smooth slow motion or restore low-frame-rate footage. Recent AI video-generation systems can produce convincing missing frames in difficult scenes, such as fast motion or temporary occlusions, but they are usually too slow for practical use because they repeatedly solve many small frame-by-frame problems. Our work introduces SpeedVFI, a way to generate all missing frames in a short video sequence at once. Instead of running a video model over and over, SpeedVFI looks at all given key frames together and produces the full smooth sequence in a single pass. It also reduces unnecessary computation by focusing most of the model’s effort on the frames that need to be created. As a result, SpeedVFI makes AI-based video frame interpolation much faster while keeping strong visual quality. This can help make high-quality video restoration, slow-motion generation, and efficient video editing more practical.