Read the manual
Watch promo video 1
Watch promo video 2
UELTX2: Unreal to LTX-2 Curated Video Generation
Generate cinematic placeholders, animated textures, and dynamic VFX locally inside Unreal Engine 5. A native bridge for Lightricks’ open-source LTX-2 model via ComfyUI. No cloud fees. 100% Private.
1. The Dynamic Asset Pipeline (VFX & Textures)
The Problem: Creating high-quality animated textures (Flipbooks/SubUVs) for fire, water, magic portals, or sci-fi screens requires complex simulations in Houdini or EmberGen, which take hours to set up and render.
The LTX-2 Solution:
Workflow: You prompt LTX-2: "Seamless looping video of green toxic smoke, top down view, 4k."
Result: You get a video file.
UE5 Integration: You automatically convert that video into a Flipbook Texture and plug it immediately into a Niagara Particle System.
Rationale: You can generate unique, specific VFX assets in 2 minutes inside the editor, rather than spending 4 hours in simulation software.
2. Rapid Pre-Visualization (Animatics)
The Problem: During the "Greyboxing" or layout phase, level designers usually put static mannequins in the scene. To visualize a cutscene or a complex event (e.g., a building collapsing), animators must create a "blocking" animation, which takes days.
The LTX-2 Solution:
Workflow: Place a 2D Plane in the level. Select it, type "Cyberpunk building collapsing into dust," and hit Generate.
Result: The plane plays the generated video.
Rationale: Directors and Level Designers can visualize timing, mood, and lighting of dynamic events zero animator hours. It allows for "Disposable Ideation"—generating 20 variations of a cutscene idea before committing to the expensive 3D production.
3. In-World Screens and Moving Backgrounds
The Problem: Creating content for TVs, holograms, or distant dynamic backdrops (like a busy city outside a window) in a game world is tedious. Rendering them in real-time 3D wastes performance (Draw Calls).
The LTX-2 Solution:
Workflow: Use Image-to-Video. Take a screenshot of your game assets, prompt LTX-2 to "animate traffic," or "add tv static and glitch effects."
Result: A video file you map to a MediaTexture.
Rationale: You get high-fidelity, "3D-looking" motion on a cheap 2D surface. This saves massive amounts of GPU frame time compared to rendering actual 3D cars in the distance or simulating complex UI glitches.
4. IP Privacy and Cost (The "Local" Rationale)
The Point: Why LTX-2 specifically, and not Veo/Runway/Sora/Kling?
Security: IP leaks are bad for you. If you use Runway/Midjourney, you are uploading concept art to a cloud server. LTX-2 runs locally. Your unannounced game assets never leave your internal network.
Cost: No subscription fees. If you iterate 500 times on a texture, it costs you electricity, not credits.
Modifiability: Since LTX-2 is open weights, you can Fine-Tune (LoRA) the model on your game's specific art style. You can train it to understand what your main character looks like, ensuring the generated placeholders actually look like your game.
Summary: The "Hybrid" Workflow
The ultimate point of this integration is the Hybrid Workflow.
Instead of:
Model -> Rig -> Animate -> Render -> Import
The workflow becomes:
Block out rough scene in UE5 -> Screenshot -> LTX-2 (Image-to-Video) -> Final Polish.
You use UE5 for structure and perspective (which it is perfect at) and use LTX-2 for detail, texture, and complex movement (which is hard to animate manually).
Additional information
🎥 The First Native LTX-2 Integration for Unreal Engine 5
Stop waiting for cloud queues and paying subscription fees for generative video. UELTX2 brings the power of Lightricks’ state-of-the-art LTX-2 (Diffusion Transformer) model directly into your Editor workflow.
This plugin acts as a native bridge between Unreal Engine 5 and your local ComfyUI instance, allowing you to generate 4K-ready video assets, moving textures, and pre-vis animatics without ever leaving the Viewport.
⚡ Use Cases
Dynamic Textures (I2V): Right-click any static texture in your Content Browser and use Image-to-Video to bring it to life. Turn a static photo of smoke into a looping flipbook source, or animate a TV screen texture in seconds.
Rapid Greyboxing: Need a cinematic of a building collapsing for a level mock-up? Generate it in 60 seconds with Text-to-Video instead of spending 3 days animating blocks.
VFX Elements: Generate "magical aura" or "fluid simulation" video files to plug directly into Niagara Media Plates.
🔒 Enterprise-Grade Privacy
Your prompt data and game assets never leave your local network. UELTX2 runs on localhost, making it safe for NDA-bound projects where uploading concept art to external cloud servers is prohibited.
🛠 Core Features
Native Editor Panel: A dockable Editor Utility Widget for streamlined prompting.
Context Menu Integration: Right-click Actions for "Animate this Texture" (I2V).
Auto-Import Pipeline: Generated .mp4 files are automatically imported as FileMediaSource assets, ready to play.
Hybrid Workflow: Uses UE5 for the UI and ComfyUI for the heavy lifting (VRAM management).
Open Architecture: Fully customizable JSON templates. Change samplers, steps, or resolutions by editing the included JSON files.
⚠️ Technical Requirements (Read Before Buying)
This plugin requires external software setup. It does not contain the AI model itself (which is 8GB+).
Hardware: NVIDIA GPU with 12GB VRAM (Minimum for Quantized GGUF) or 24GB VRAM (Recommended for Full Precision).
Backend: You must have ComfyUI installed locally.
Model: You must download the LTX-2 weights (free via HuggingFace).
OS: Windows 10/11 (Linux not supported in v1.0).
📦 What's Included
UELTX2 Plugin: C++ Runtime & Editor Modules.
Workflow Templates: Optimized .json workflows for LTX-2 Text-to-Video and Image-to-Video.
Documentation: Comprehensive HTML guide on setting up the local server and acquiring the model weights.
⚖️ Legal Notice
Plugin License: You are purchasing the UELTX2 Bridge Plugin code (Proprietary via Fab).
Model License: The LTX-2 Model weights are developed by Lightricks and are subject to their Open Access License. Users are responsible for adhering to the model's usage terms.
(c) Andras Gregori @ GregOrigin 2026