SolarWM goes fully open-source: 1.43M clips, 25TB pipeline for long-horizon video world models
Scobleizer · x · 2026-09-03
A team from CUHK-SZ, NUS, NVIDIA, MSRA and others has fully open-sourced SolarWM, an end-to-end pipeline for video world model research, which Scobleizer called "accelerant for robotics."
- SolarWM-Data: 1.43M video clips (25 TB) with camera annotations, quality scores, and the full curation/filtering stack.
- Long-horizon interaction: SolarWM-5B, trained only on 5-second clips, achieves real-time interactive generation with stable rollouts spanning minutes to hours.
- One framework, 4 backbones: unified training/inference for SolarWM-Wan-5B, Wan-14B, LTX-2.5, and Minimax-H3, including a causal student variant (4 steps).
Paper, code, checkpoints and datasets are all released as turnkey infrastructure for the community.
Related event: CUHK-SZ and NVIDIA, NUS, MSRA Fully Open-Source SolarWM Video World Model(5 posts)→
More from Embodied
- 704-vote robotics Elo ranks Claude Opus 5.5 first at 1379; no model completes the task — qinzytech · 2026-10-21
- Walden Robotics CTO joins Long Horizon panel on large behavior models with Rodney Brooks — adnothing · 2026-10-21
- YC-backed Antropi Robotics emerges with autonomous CNC factories, already profitable in 6 months — ycombinator · 2026-10-21
- Why the $3,000+ iPhone Duo is sold out until January: one buyer's theory — RayFernando1337 · 2026-10-21
- Robotics startup launches Deployments to serve fine-tuned HF checkpoints to robots in real time — arnie_hacker · 2026-10-21
- Amazon FAR's FIBER uses action-aligned latent planning to let Unitree G1 open doors and carry boxes — ChongZzZhang · 2026-10-21