Introducing LTX-2.5: The Open World Model for Video, Real-Time, & Physical AI

The latest model is a jump step improvement across quality, efficiency, and speed, while keeping users in full control.

Today we are releasing LTX-2.5, the most capable open weights world model on the market. We built it to provide an even more performant foundation across sectors where LTX models are already in production, including film, robotics, and real-time world generation. LTX-2.5 is available to download and run anywhere today with open weights, to empower people to build with full control.

How LTX-2.5 Measures Up

We believe that open weights are the best way to enable people to build with full control across the hardware they use, model customization, and control of their IP. This is why builders are already widely using LTX models in production for film, robotics, and real-time rendering. With the release of LTX-2.5, we’re working with Asteria, ComfyUI, Markov Robotics, and Reactor, each pushing the boundaries of what’s possible with world models in their fields. LTX is already the most used open world model, with more than 33 million downloads. LTX-2.5 further extends our leadership, with significant improvements across quality, efficiency, and speed.

“World models face challenges that LLMs never had to solve, like holding motion, space, and sound consistent across time, which is why efficiency and control matter so much. By keeping LTX open, we let teams own their hardware, their IP, and their model. LTX-2.5 is the most capable model yet, and partnerships like the one with ComfyUI, built on a shared belief in an open future, are how it reaches the teams that need both industry-leading quality and full control.” — Zeev Farbman, co-founder & CEO, LTX

Through our strategic launch partnership with ComfyUI, all of LTX-2.5 is available from day one inside the node-based environment the open community already builds in, from a first experiment to a standardized production pipeline.

“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. Open is what lets the community move fast and lets businesses build on what it proves.” — Yoland Yan, co-founder & CEO, ComfyUI

What’s New in LTX-2.5

For LTX-2.5, we’ve rebuilt nearly every stage of the generation pipeline rather than bolting capabilities onto an older core for:

• Professional-grade output. A new diffusion video decoder that reduces visual artifacts in high motion while keeping the high compression ratio of LTX, and remaining true to existing footage.
• Consistency across sequences. Native multishot generation renders a full sequence as one output, holding character, scene, and voice across cuts, while a custom Gemma 4 language backbone and dedicated prompt enhancer read and understand complex, multi-subject prompts more accurately.
• High-fidelity where it counts. Diffusion Fidelity Rendering builds motion and structure in an 8x temporally compressed latent space while generating high-fidelity keyframes to anchor visual detail, adapting the number of keyframes to scene complexity and compute budget, so pixel-level quality lands where it matters without full render cost on every frame.
• Domain adaptation. A pretrained checkpoint tuned for physical AI and robotics gives teams a flexible base to fine-tune on their own domain data, so builders can adapt the model to a world that looks nothing like cinematic video.
• Efficiency at scale. A substantially better distilled model delivers the same quality at lower cost and faster inference, so teams can run near-full-model quality on hardware they already own and deploy it economically at production volume.
• Local optimization. Together with NVIDIA, we optimized LTX-2.5 for local inference on NVIDIA RTX GPUs and DGX Spark, cutting memory requirements so more teams can run a frontier world model on hardware they already own.

“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.” — Gerardo Delgado Cabrera, Senior Director of Product for Local AI, NVIDIA

The result is a sharper, more physically accurate output, generated at a fraction of the cost and render time of closed alternatives. Because LTX-2.5 is open weights, you run it on your own hardware, so your data and IP never leave your devices, and you can fine-tune and extend it for whatever your work demands.

The Next Infrastructure Layer of AI

Large language models (LLMs) learned to predict the next word. World models learn to predict the next moment. They generate these environments, simulate how they behave, and let you act inside it. They are becoming the foundation for everything from film and advertising to gaming, simulation, and physical AI, supporting robots as they move through a warehouse or factory floor, operating within an offshore rig, the power grid, and so much more. With it, we’re establishing one open model across worlds, characters, audio, and motion, that extends into physical action, and providing innovative companies across media and entertainment, real-time applications, and robotics, access to the foundation for world models they can run within their own environments today.

Built for Real Production Workflows

An open model is only as strong as what people build on it, and our proof is that teams are already shipping on LTX across three very different frontiers today. In the media and entertainment industry, studios like Asteria produce original film and video on LTX. Go deeper on LTX for media & entertainment in our companion post.

For the development of real-time applications, we’ve partnered with Reactor, the developer platform for real-time generative video. Reactor runs LTX-2.5 on its low-latency inference infrastructure to power interactive avatars, live worlds, and real-time robotics workloads, giving developers the ability to build production-grade real-time experiences without building that infrastructure themselves.

“The future of media is real-time, and it only works if the underlying model is fast, controllable, and yours to run. Partnering with LTX means we can tune the model to our own infrastructure and bring live, responsive avatars to users worldwide.” – Alberto Taiuti, co-founder and CEO, Reactor

In robotics, Markov Robotics uses LTX to develop how physical systems perceive and move through the world, working to solve the generalization problem this field is currently facing.

“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. We have not found another open model that does this for us the way this one does.” — Atharva Gundawar, co-founder and CEO, Markov Robotics

Where This Goes Next

We believe world models will run behind a large share of what the world watches, plays, and automates over the next decade, and we intend to build that future in the open, with the people using these models on real problems.

LTX-2.5 is available today as open weights on Hugging Face, natively in ComfyUI, and through the LTX API for teams that want managed, production-grade generation. It runs on hardware you already own, whether a data center GPU to a Mac, and it’s free to use for organizations under $10M in annual recurring revenue. We look forward to seeing what you build.