Image To Video Temporal Extension

via “video generation and frame interpolation with temporal consistency”

🤗 Diffusers: State-of-the-art diffusion models for image, video, and audio generation in PyTorch.

Unique: Uses temporal attention layers that compute cross-frame attention, enabling the model to enforce consistency across frames without explicit optical flow or motion estimation. Unlike frame-by-frame generation, temporal attention allows the model to learn smooth motion trajectories and prevent flickering by attending to neighboring frames during denoising.

vs others: More efficient than frame-by-frame generation with optical flow because it avoids explicit motion estimation and stitching, instead learning temporal coherence end-to-end. Outperforms simple frame interpolation because it generates novel content rather than blending existing frames.