Intro Video

Frame In-N-Out is a controllable Image-to-Video generation Diffusion Transformer model where objects can enter or exit the scene along user-specified motion trajectories and ID reference. Our method introduces a new dataset curation pattern recognition, evaluation protocol, and a motion-controllable, identity-preserving, unbounded canvas Video Diffusion Transformer, to achieve Frame In and Frame Out in the cinematic domain.

Model Zoo 🤗

📚 Citation

@article{wang2025frame,
  title={Frame In-N-Out: Unbounded Controllable Image-to-Video Generation},
  author={Wang, Boyang and Chen, Xuweiyi and Gadelha, Matheus and Cheng, Zezhou},
  journal={arXiv preprint arXiv:2505.21491},
  year={2025}
}