Direct the motion, camera, and structure of your AI video.
Replace prompt roulette with locked control. Guide motion with trajectory splines, lock a camera move, or hold the structure of a reference. Set it once and reuse it across takes. Open weights, self-hosted.
.svg)
Key capabilities
Motion Track Control: Guide motion with splines
Guide object or region motion with point-trajectory splines from SpatialTrackerV2 or ComfyUI, on LTX-2.3-22B. Reference runs at 0.5x output resolution.
Camera Control: Lock a camera move
Load a pre-trained camera-move LoRA on LTX-2 (19B) to lock dolly in, dolly out, dolly left, dolly right, jib up, jib down, or static. No training required.
Repeatable motion choreography
Guide object motion with drawn or tracked splines once, then reuse that path across multiple generations. No re-rolling to get the same move twice.
Locked camera moves
Load a pre-trained camera-move LoRA and lock in dolly, jib, or static motion. Compose and reuse the move across a campaign or sequence.
Structure-conditioned generation
Match the pose, depth, or edges of a reference image or frame while generating new content. Keep compositional control while letting the model fill the details.
Directable generation in production
Self-host the full pipeline with repeatable motion, camera, and structure. No manual intervention between takes, no prompt roulette.
Built for
VFX and animation teams whose high-volume shot production demands repeatable motion and camera moves across sequences and campaigns. Generative AI leads building production pipelines with ControlNet-style conditioning for directable, scalable video generation.
Product teams, AI startups, and developers building AI-powered video features. Add production-grade video generation as a product capability, not a research project. One API, production-ready results, and no custom orchestration.
Brands, agencies, and creative teams producing high volumes of content. Turn existing assets into video at scale. Faster iteration, lower production cost, and more output from what you already have.
Teams that require full control over deployment and data. Run video generation in your own environment. On-premises, no cloud dependency, and full infrastructure ownership.
Platforms powering creative tools with multiple AI models. Upgrade your video output with a best-in-class engine. Improve generation quality, retain users, and differentiate with a model built for production, not prototypes.
How it works
Input
Technical characteristics:
- Motion Track — Reference video with colored trajectory splines from SpatialTrackerV2 or the ComfyUI trajectory node. Base: LTX-2.3-22B plus ltx-2.3-22b-ic-lora-motion-track-control-ref0.5.safetensors. Reference runs at 0.5x output resolution (downscale factor 2).
- Camera — Load a pre-trained camera-move LoRA directly on the LTX-2 (19B) checkpoint. No reference or training needed.
- Union — Reference processed as Canny, depth, and/or pose. Base: LTX-2.3-22B plus Union Control IC-LoRA. Reference runs at 0.5x output resolution (downscale factor 2). Optional init image for image-to-video.
- Pipeline — Run through the IC-LoRA reference-conditioning workflow in the ComfyUI-LTXVideo repo.
Output
Technical characteristics:
- Generated clip — Follows the supplied motion paths, camera move, and/or structural control.
- Framework — Motion Track and Union are built on the AVControl framework.
- Self-host — Official workflows in ComfyUI-LTXVideo on GPU, under the LTX-2 Community License.
What users are building with LTX
FAQs
How do I control motion?
Supply a reference video with point-trajectory splines, extracted with SpatialTrackerV2 or drawn in the ComfyUI node. The model follows those paths.
What camera moves are available?
Seven pre-trained camera-move LoRAs on LTX-2 (19B): dolly in, dolly out, dolly left, dolly right, jib up, jib down, and static. Load them directly with no training needed.
What structural controls are supported?
Union Control handles Canny (edge), depth, and pose conditioning in a single IC-LoRA on LTX-2.3-22B.
Do I need full-resolution control input?
No. For Motion Track and Union, the reference runs at 0.5x output resolution (downscale factor 2).
How do I run it?
Open weights on HuggingFace with official workflows in ComfyUI-LTXVideo under the LTX-2 Community License.
