FL4NFeed-forward Latent Model for 4D Novel View Synthesis

4D novel view synthesis renders a dynamic scene from viewpoints and instants that were never observed. Feed-forward methods do it with explicit 3D Gaussians plus motion, and pay with ghosting, floaters and holes.

A frozen static latent backbone supplies the 3D prior; a lightweight dynamic encoder adds a temporal residual on top. Training needs only images and binary motion masks.

4D Novel View Synthesis (NVS) from monocular videos aims to render dynamic scenes across viewpoints and timestamps, requiring jointly inferring 3D structure and temporal dynamics. Existing feed-forward methods rely on explicit 3D Gaussian representations, which are prone to temporal inconsistency and severe rendering artifacts. We present FL4N, the first feed-forward latent 4D NVS model. We bypass explicit 3D primitives by weaving a geometrically grounded scene prior into a unified latent representation across space and time. To overcome dynamic training data scarcity, we learn a temporal residual on top of a frozen, pretrained static latent backbone. A lightweight dynamic encoder learns to isolate true scene motion from camera parallax. Trained purely on image synthesis with binary mask guidance, FL4N achieves state-of-the-art rendering quality for 4D NVS in real time. Our proposed latent representation successfully learns a strong 4D prior, generalizes across novel views and timestamps, and cleanly decouples view and time in the latent space.