LTX has released LTX-2.5, an open-weights world model that generates video and simulates how physical environments change—capabilities increasingly important for robotics and AI systems.
The model is free to use locally on developer hardware through Hugging Face and ComfyUI, allowing organizations to maintain control over their data and customize the system for specific applications like robot training and real-time video generation.
LTX's world models have been downloaded more than 33 million times and are already in use across film production, robotics, and real-time rendering.
What happened
LTX released LTX-2.5, an open-weights world model designed to generate video and simulate how physical environments change over time. The model is available free through Hugging Face and ComfyUI, with native support for Nvidia RTX GPUs and the ability to run on local hardware from data centers to Macs. LTX's world models have now recorded more than 33 million downloads.
Why it matters
World models are increasingly used in robotics to help systems predict how environments and objects behave—a capability that goes beyond what language models do. By keeping the model open and optimized for local hardware, LTX allows developers and organizations to run, customize, and fine-tune it on their own infrastructure without relying on cloud services, protecting proprietary data and intellectual property. This matters for robotics companies, filmmakers, and enterprises building real-time applications.
What to watch
LTX-2.5 is free for organizations with less than $10 million in annual recurring revenue. The model includes a pretrained checkpoint specifically tuned for physical AI and robotics, which developers can fine-tune with their own domain-specific data. Companies including Markov Robotics, Reactor, and Asteria are already using LTX technology for robotics applications and film production.
LTX has launched LTX-2.5, its latest open-weights world model, which the company says delivers improvements in visual quality, prompt understanding, generation speed, and efficiency while enabling developers and enterprises to run and customize the model on their own hardware. The model is available immediately through Hugging Face, with native support in ComfyUI and access through the LTX API for managed deployments. LTX reports that its world models have now recorded more than 33 million downloads and are being used in film production, robotics, and real-time rendering applications.
Unlike incremental updates, LTX rebuilt much of the generation pipeline for LTX-2.5. Key architectural additions include native multishot generation, which allows the model to generate a complete video sequence while maintaining consistency between characters, scenes, and voices across cuts. Prompt understanding has been upgraded through a custom Gemma 4 language backbone and a dedicated prompt enhancer designed to interpret complicated prompts involving multiple subjects. For visual output, LTX introduced a new diffusion video decoder intended to reduce artifacts during high-motion sequences while retaining the model's high compression ratio. The company also developed Diffusion Fidelity Rendering, which constructs motion and structure in an 8x temporally compressed latent space while generating higher-fidelity keyframes to preserve important visual details. The number of keyframes adjusts according to scene complexity and available computing resources, concentrating processing on parts of sequences where greater detail is required. Improvements to the distilled model enable near-full-model quality with lower computing requirements and faster inference.
A major focus for LTX-2.5 is physical AI and robotics. World models can help robots learn how environments behave and predict how actions affect their surroundings—a capability distinct from conventional video generation. LTX-2.5 includes a pretrained checkpoint specifically tuned for physical AI and robotics, which developers can fine-tune using domain-specific data. Atharva Gundawar, co-founder and CEO of Markov Robotics, stated: "Training robots means teaching them how the physical world actually behaves, not just what it looks like. LTX-2.5 is the open model that gets closest to that for us, and being able to run and fine-tune it on our own hardware is what makes it usable for real robotics work." A world model attempts to predict the next state or "moment" within an environment—fundamentally different from a language model's word-prediction task—making the technology applicable to simulation, robot training, planning, and conventional media applications.
LTX has partnered with Nvidia to optimize LTX-2.5 for local inference on Nvidia RTX GPUs and DGX Spark systems, reducing memory requirements so more developers and businesses can run the model locally rather than relying entirely on cloud infrastructure. Gerardo Delgado Cabrera, senior director of product for local AI at Nvidia, said: "Local models enable creators and developers to freely explore their ideas thanks to their local GPUs. Through close work with LTX to optimize LTX-2.5 and enable it in popular frameworks like ComfyUI, users with Nvidia RTX GPUs and DGX Spark can get the best performance with significantly less VRAM consumption." LTX says LTX-2.5 can run on hardware ranging from data center GPUs to Macs, allowing organizations to keep proprietary data and intellectual property on their own infrastructure. LTX has also partnered with ComfyUI to make LTX-2.5 natively available within the node-based development environment from launch, allowing developers to move from experimentation to production workflows without transferring work to a separate proprietary platform. Yoland Yan, co-founder and CEO of ComfyUI, said: "ComfyUI and LTX share a simple belief, that open models build better tools, faster. With LTX-2.5 running natively in ComfyUI, a developer can go from an idea to a working world-model pipeline in an afternoon, and an enterprise can take that same pipeline straight into production on its own hardware." LTX is also working with Reactor, a development platform for real-time generative video, running LTX-2.5 on its low-latency inference infrastructure for interactive avatars, live generated environments, and real-time robotics workloads. Alberto Taiuti, co-founder and CEO of Reactor, stated: "The future of media is real-time, and it only works if the underlying model is fast, controllable, and yours to run." Asteria is using LTX technology for film and video production. LTX-2.5 is free to use for organizations with less than $10 million in annual recurring revenue.
LTX-2.5 represents a significant shift in how video generation and physical simulation are being democratized. Rather than adding incremental features to its previous architecture, LTX rebuilt much of the generation pipeline, introducing multishot generation (which maintains consistency between characters, scenes, and voices across cuts), improved prompt understanding through a custom Gemma 4 language backbone, and a new diffusion video decoder designed to reduce artifacts during high-motion sequences. The company has also developed Diffusion Fidelity Rendering, which constructs motion and structure in a compressed latent space while generating higher-fidelity keyframes only where visual detail matters most—allowing the system to concentrate computational resources based on scene complexity.
The robotics and physical AI angle is particularly noteworthy. By including a pretrained checkpoint specifically tuned for physical AI and robotics, LTX is positioning world models as infrastructure not just for media creation but for systems that must learn how the physical world behaves. Markov Robotics's statement that "training robots means teaching them how the physical world actually behaves" underscores why this differs from traditional video generation: robots need models that predict state transitions in environments, not just visually plausible sequences. The ability to fine-tune the model locally on proprietary hardware addresses a critical pain point—keeping sensitive robotics data and intellectual property on-premises rather than sending it to cloud services.
The partnership with Nvidia, ComfyUI, and Reactor also signals that LTX is building an ecosystem where developers can move from experimentation to production without switching platforms or surrendering control to proprietary systems. The free-to-use tier for organizations under $10 million in annual recurring revenue, combined with open-weights release, removes traditional barriers to adoption.
AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.
Free · takes 30 seconds · unsubscribe anytime
Ask AI anything about this article. Q&As are published on this page for other readers too.
Foxconn announced on August 12 at its second-quarter 2026 earnings conference that it is broadening its busine…
South Korea plans to establish a strategic investment account with at least 20 trillion won in assets within t…
Nvidia CEO Jensen Huang signed seven Japanese companies—Fujitsu, FANUC, Yaskawa Electric, Kawasaki Heavy Indus…
Meta is simultaneously launching open-weight AI models like Muse Spark 1.2 and Muse Glimmer, expanding data-ce…
Taiwan's manufacturers are expanding into humanoid robot supply chains as global demand for robotics rises, ca…
DeepSeek released V4 Pro 0813, its latest Pro model, available through OpenRouter via API
The AI news that matters, in one minute each morning.
Sign up free