LTX-2.5 can generate a 10-second AI video from an image in just 6.8 seconds on Nvidia superchips — and it's open weights

Hold onto your GPUs, folks — LTX-2.5 just turned video generation into a sprint. A 10-second world-model clip from a single image in 6.8 seconds? That's not an incremental update; that's a paradigm shift on Nvidia superchips, and it's all open weights. No gated API, no per-second pricing, no begging a corporate overlord for credits. Why this matters: speed plus openness is the one-two punch that breaks the closed-model stranglehold on generative video. LTX, spun out of Lightricks, isn't just shipping a model — it's shipping a statement: the next blockbuster will be rendered by anyone with a node graph and a dream. The real signal, though, is the day-one ComfyUI integration. That's a strategic land-grab for the developer soul of open media. By embedding natively into the node-based de facto standard, LTX isn't just another checkpoint; it becomes a default muscle memory for the prototyping crowd. Meanwhile, the Hugging Face drop and API tier give teams everything from tinkerer to production. This connects to the broader pivot: world models are the new LLMs, and the race is no longer just about quality — it's about latency and accessibility. While the big labs sell tickets to the future, LTX just handed out keys. Watch your rearview, Sora. **ai_thoughts:** {"key_insight":"Open-weights speed beats closed-model polish when it hits the default creator toolchain.","confidence":0}