Tech With Tim

This Open-Source AI Video Model Just Crushed Sora

The video introduces LTE S2, an open‑source diffusion‑transformer hybrid that generates synchronized video and audio. Released by LTA, the model can be run locally on a single GPU with as little as 12 GB of VRAM, delivering native 4K output and clips up to 30 seconds long. According to the presenter, LTE S2 is up to 18 times more efficient than the previous state‑of‑the‑art model (e.g., Sora), completing a 10‑second clip in roughly 15‑20 seconds on an RTX 3060. Benchmarks on data‑center H100 GPUs show 49 steps versus 2.69 steps per minute for competitors. The full repository, including 300 GB of weights, training code, and ComfyUI workflows, is publicly available. Community members have already produced impressive results, from lip‑synced dialogues to 27‑second continuous scenes, all generated on consumer hardware. The creator highlights integrations such as a Premiere Pro plugin and control adapters for motion, camera behavior, and LoRAs, demonstrating the model’s flexibility beyond a simple API. By removing proprietary barriers, LTE S2 turns video‑AI into a usable infrastructure for studios, VFX teams, and solo creators who need on‑premise control, unlimited generation, and seamless pipeline integration. Its efficiency and open nature could accelerate adoption of generative video across content production and lower costs dramatically.