# LTX Releases Open-Weight LTX-2.5 World Model > Source: > Published: 2026-08-11 14:56:12+00:00 # LTX Releases Open-Weight LTX-2.5 World Model LTX released LTX-2.5, an open-weight video and world model, on August 11, with native availability in ComfyUI and downloadable weights on Hugging Face. VentureBeat reports that the release adds a diffusion video decoder, native multishot generation, and new conditioning modes. CNET reports that the model is optimized for local inference on Nvidia RTX GPUs, supporting local customization for video production and robotics workflows. LTX released **LTX-2.5**, an open-weight video and world model, on August 11. The model is available through Hugging Face, ComfyUI, and the LTX API, according to VentureBeat. The company launched the release with native integration in ComfyUI, the node-based generative-media workflow environment. LTX-2.5 targets video generation and physical-AI use cases in which a model must maintain visual and spatial consistency over time. CNET describes world models as systems intended to represent aspects of physical environments for simulation, content generation, and robot development, rather than primarily generating text as large language models do. ### Video pipeline changes According to VentureBeat's account of the announcement, LTX-2.5 rebuilds much of its generation pipeline. Reported additions include a diffusion video decoder intended to reduce artifacts in high-motion footage and reconstruct fine details such as faces and text, plus native multishot generation for producing multiple shots as one sequence. VentureBeat also reports a custom Gemma 4 language backbone, a prompt enhancer, new conditioning modes, and improved support for autoregressive models. LTX co-founder and CEO Zeev Farbman told VentureBeat that the release adds "multi-shot support, a diffusion decoder for better quality, new conditioning modes, better support for autoregressive models that are critical for real-time use cases and robotics." VentureBeat reports that LTX-2.5 can generate a 10-second image-to-video clip in 6.8 seconds on Nvidia superchips. That result is a reported performance figure rather than an independently reproduced benchmark in the available coverage, so practitioners evaluating throughput would need to compare hardware configuration, resolution, sampling settings, and output quality before treating it as a deployment baseline. ### Open weights and local execution CNET reports that LTX-2.5 is optimized for local inference on Nvidia RTX GPUs, allowing developers to run the model on their own hardware. VentureBeat reports that the weights are free for organizations below $10 million in annual recurring revenue, while larger companies require a negotiated license. The licensing distinction matters because "open weights" does not necessarily mean unrestricted use. Teams considering self-hosted generation should review the applicable license terms, model dependencies, GPU memory requirements, and any restrictions on commercial deployment before integrating the weights into a production pipeline. LTX claims its model family has surpassed **33 million downloads**, a figure reported by VentureBeat and Techzine. Techzine also reports that LTX is working with Asteria, ComfyUI, and Reactor. The available reports do not provide independently audited usage figures or comparative quality benchmarks against closed video models. ### Relevance for video and robotics builders For generative-video teams, native ComfyUI support could reduce the integration work required to test prompting, conditioning, and multi-stage workflows. For robotics and simulation researchers, the relevant question is less whether a model generates visually plausible clips and more whether its temporal, spatial, and physical representations remain reliable under varied scenes and actions. Across comparable open-model releases, local weights can give teams more control over fine-tuning data, inference infrastructure, and intellectual-property handling. They also transfer evaluation, serving, safety controls, and model-update responsibilities from a hosted provider to the deploying organization. ## Key Points - 1LTX-2.5 combines open weights, local Nvidia RTX inference, and ComfyUI access, expanding options for self-hosted video-generation experimentation. - 2Reported multishot generation and diffusion decoding target temporal consistency and visual artifacts, two persistent evaluation challenges in generative video. - 3Comparable open-weight deployments provide customization control but shift licensing review, infrastructure operation, evaluation, and safety governance to adopters. ## Scoring Rationale LTX-2.5 is a notable open-weight release for teams building controllable video-generation, simulation, and physical-AI workflows. Its ComfyUI integration and local-inference positioning are practically relevant, although the available coverage does not include independent benchmark results or detailed robotics validation. ## Sources Primary source and supporting public references used for this report. Practice interview problems based on real data 1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with. [Try 250 free problems](/problems)