LTX released LTX-2.5, an open-weights video generation model that produces 10-second videos from static images in 6.8 seconds when running on Nvidia's latest superchips. The model launches with native integration into ComfyUI, the industry-standard node-based workflow tool for generative media development.
The release marks a shift toward accessible video synthesis. LTX, a spinoff from Lightricks, designed LTX-2.5 as a "world model" that understands spatial and temporal coherence rather than simply predicting pixel sequences. This architectural choice enables faster generation and better consistency across frames.
The weights are available immediately on Hugging Face and through the LTX API for teams preferring managed infrastructure. Pricing follows a clear structure: free for organizations with under $10 million in annual recurring revenue. Larger enterprises negotiate custom licenses.
ComfyUI integration on day one matters. The platform has become the de facto prototyping environment for creators and researchers building with open generative models. Including LTX-2.5 directly within ComfyUI's node-based interface eliminates friction. Users can chain video generation into existing workflows without context-switching or API integrations.
Speed matters for adoption. At 6.8 seconds to generate a 10-second clip on Nvidia H100 or H200 chips, LTX-2.5 approaches interactive latency. Competitors like Runway and Pika typically require 30 seconds or longer for similar outputs. This speed opens practical applications: rapid iteration during creative work, real-time previews in video editors, and batch processing at scale.
The open-weights strategy differentiates LTX from closed models. Researchers can fine-tune, audit, and customize the model for specific use cases. Organizations concerned about API dependency