LTX-2.5 - LTX's open-source AI video generation foundation model

What is LTX-2.5

LTX-2.5 is LTX's open-source AI video generationfoundation model with 22B parameters, released with open weights upon launch. The model supports native multi-shot generation, 4K HDR and RAW/EXR professional workflows, can generate audio synchronously, and has precise video editing capabilities. At a time when closed-source models are increasingly expensive, LTX-2.5 provides a professional-grade open-source alternative for the AI video field through an open ecosystem strategy.

Main features of LTX-2.5

• Native multi-shot generation: A single generation can output multiple continuous shots from different camera positions and shot sizes, maintaining a high degree of consistency in character appearance, environmental atmosphere, lighting conditions, and sound across shot changes, solving the core pain point of shot continuity in AI narrative creation.
• Diffusion fidelity rendering: Adaptively allocates rendering compute according to scene complexity, with complex dynamic shots receiving more computing resources while static or simple scenes automatically save resources, improving overall efficiency while ensuring image quality.
• Precise video editing: While preserving the motion trajectory and spatial structure of the original live-action footage, it performs high-precision style rewriting, scene repainting, and character replacement, enabling controllable creation of "live-action skeleton + AI stylized overlay".
• Automatic duration prediction: A built-in Duration Head module reads the action description in the prompt and automatically determines the most suitable clip length, with no need for creators to experiment repeatedly.
• Professional-grade output formats: Supports native 4K HDR, RAW, and EXR lossless format output, and can be directly integrated into professional film and television color grading, visual effects compositing, and post-production editing pipelines.
• Synchronous audio generation: Through an independent Audio VAE module, it synchronously generates a matching sound track for the video.
• IC-LoRA precise control: Provides three reference video control modes—Canny edges, Depth, and Pose—so that generation results strictly follow the composition, spatial structure, or character actions of the reference material.

Technical principles of LTX-2.5

• Gemma 4 12B text encoder: A large language model encoder customized by Lightricks specifically for LTX, capable of fully preserving multi-subject relationships, action details, lighting descriptions, and camera direction in complex long prompts, without losing clause information as prompts become longer.
• Prompt Enhancer: A lightweight auxiliary model that automatically expands short user prompts into detailed cinematic instructions, lowering the barrier to prompt engineering with extremely low computational overhead.
• Diffusion Video Decoder: A new architecture that replaces the traditional VAE decoder, specifically optimized for fast-motion scenes, significantly reducing smearing and artifacts so that faces are clearer and text is more readable.
• Audio VAE: An independent audio latent space encoder responsible for mapping video content into matching audio representations, enabling synchronous generation of visuals and sound.
• Diffusion Fidelity Rendering: Adopts a two-stage rendering strategy of "structure first, details later". The first stage builds the motion, composition, and shot framework in an 8x temporally compressed latent space, and adaptively generates high-fidelity keyframes; the second stage jointly diffusion-renders the final video from the structure and keyframes, resolving textures, materials, and facial details with pixel-level precision.
• Duration Head: A prediction module deployed before the diffusion process that, by understanding the action semantics in the prompt, automatically infers and outputs the most suitable video duration, giving the model a preliminary perception of narrative rhythm.

Follow WeChat and reply "open source", joinAI Open Source Project Communication Group

How to Use LTX-2.5

• Choose a Path: Decide whether to use the API cloud service or local deployment with ComfyUI.
• API Access: Apply for an API Key on the official LTX website, and choose the Fast or Pro tier to call the interface.
• Install Nodes: For local deployment, update ComfyUI to the latest version and install the ComfyUI-LTXVideo node package via Manager.
• Download Models: Search for “LTX-2.5″ in the ComfyUI template browser, select the corresponding workflow, and download all model files with one click.
• Configure Parameters: Set width and height to multiples of 32, frame count to a multiple of 1+8, frame rate to 24/25/48/50fps, and the base resolution to half the target to enable 2× upscaling.
• Write Prompts: Use a coherent passage of prose to describe the shot size, scene lighting, character actions, camera movement, and audio; avoid tag stacking or weight syntax.
• Run Generation: Click Run. Text-to-video produces the video directly; image-to-video requires uploading the first frame; first-and-last-frame mode requires uploading the first and last frames.
• Precise Control: Advanced users can load IC-LoRA in Canny, Depth, or Pose mode and connect a reference video to lock composition or motion.
• Model Selection: Use the distilled model for rapid iteration, and the full dev model for final output to obtain finer details.

Core Advantages of LTX-2.5

• Narrative Coherence: Native Multishot multi-shot generation capability; a single output can maintain a high degree of consistency in character appearance, environmental atmosphere, lighting conditions, and sound, fundamentally solving the biggest pain point of shot continuity in AI story creation.
• Leap in Controllability: Precise video editing preserves the motion trajectories and spatial structure of the original live-action footage. Combined with the three IC-LoRA control modes—Canny/Depth/Pose—it enables a controllable production shift from “blind gacha” to “live-action skeleton + AI stylized overlay.”
• Industrial-Grade Workflow: Native RAW/HDR/EXR output can be directly integrated into professional color grading, VFX compositing, and post-production editing pipelines without compression or transcoding, meeting the film and television industry's rigid demand for lossless footage.
• Adaptive Efficiency: Diffusion Fidelity Rendering technology dynamically allocates rendering compute based on scene complexity: complex dynamic shots are rendered in fine detail, while simple static scenes automatically save resources, improving overall generation efficiency while ensuring pixel-level image quality.

LTX-2.5 Project Addresses

• Official Project Website：https://ltx.io/model/ltx-2-5
• HuggingFace Model Repository：https://huggingface.co/Lightricks/LTX-2.5

Comparison of LTX-2.5 with Similar Competitors

LTX-2.5 Application Scenarios

• Professional film and television post-production: Native 4K HDR, RAW, and EXR lossless output can be directly integrated into color grading, VFX compositing, and editing pipelines, meeting the post-production needs of theatrical and streaming-level projects without suffering the color grading nightmare of compressed MP4.
• AI short dramas and animated dramas: Multishot multi-shot generation outputs coherent narrative storyboards in one go, and precise video editing supports "live-action motion skeleton + AI character replacement and stylized overlay," fundamentally solving the two fatal weaknesses of shot continuity and motion controllability.
• Advertising and brand content: Rapidly iterate high-quality commercial videos, use Prompt Enhancer to lower the barrier to prompt engineering, and rely on the Distilled model for fast previews and the full model for final output, balancing efficiency and quality.
• Physical AI and robotics: Physical AI Checkpoint provides a world model foundation for embodied intelligence teams, helping robotic systems perceive the physical world and laws of motion, going beyond the scope of pure content creation.