CMU Open-Sources Lift4D: Generating Complete Dynamic 4D Assets from a Single Video
CSProfKGD · x · 2026-07-31
Researchers at Carnegie Mellon University (CMU) have released the code for Lift4D. Accepted to SIGGRAPH Asia, the project introduces a method to generate and reconstruct complete dynamic 4D assets from a single in-the-wild video.
The pipeline operates in three stages:
- Segment: Uses SAM 3 with a text prompt to segment the target object from the video.
- Causal SAM3D: Reconstructs the object per-frame with causal latent propagation to ensure temporal coherence.
- Lift4D Training: Fuses the per-frame 3D reconstructions into a deformable 4D model.
More from Multimodal
- AI Video Editing Leaderboard: MiniMax-H3 and Gemini Tie for #1 — ArtificialAnlys · 2026-07-31
- Google's Gemini Omni Flash Debuts at #1 on Video Editing Leaderboard — ArtificialAnlys · 2026-07-31
- World's First AI-Only FAST Channel Launches in August with Over 100 Creators — Uncanny_Harry · 2026-07-31
- The AI Influencer Bottleneck: Face Inconsistency in Mid/Long Shots for LoRA Training — Fockker · 2026-07-31
- Testing MiniMax H3: Lip Sync and Ambient Sound Generation with Seed Audio — DavidmComfort · 2026-07-31
- Image Model Flux 3 Released, Receives Positive Feedback for Quality — Mr_AllenT · 2026-07-31