Multi-shot video generation, native to LTX-2.5
LTX-2.5 generates connected, multi-shot scenes in a single pass, holding character, environment, lighting, and visual style across every cut. Open weights, self-hosted or API.
.svg)
Key capabilities
Native multi-shot generation
Generate a full sequence of connected shots in one pass, not a set of clips stitched together after the fact. One prompt can span several connected scenes.
Continuity across cuts
Hold character identity, environment, lighting, voice, and visual style steady from shot to shot, without re-prompting or re-uploading references.
Storyboard to sequence
Turn a multi-scene script or storyboard into one connected sequence, generated in a single pass instead of stitched together shot by shot.
Series and short-form narrative
Keep the same character consistent across different angles and cuts within one generated sequence, reducing the continuity work required in post.
Brand and product campaigns
Hold one product, one environment, and one visual style across an entire sequence, shot to shot.
Self-hosted pipelines
Run multi-shot generation on infrastructure built for compliance, cost, or IP requirements.
How it works
Input
Technical characteristics:
- Prompt — Describe the scene as one flowing prompt: setting, sequence of events, characters, camera moves, and sound.
- Text encoder — A fine-tuned Gemma 4 12B text encoder and custom prompt enhancer parse the multi-shot instructions.
- Base — Built on the LTX-2.5 22B parameter foundation model.
Output
Technical characteristics:
- Multi-shot sequence — A connected multi-shot sequence that holds character, environment, lighting, and voice across every cut, generated in a single pass.
- Run it — Self-host via ltx-pipelines or ComfyUI, or call it through the API, under the LTX Community License.
What users are building with LTX
FAQs
What is native multi-shot generation?
LTX-2.5 generates multiple connected shots in a single pass. Other video models produce one continuous shot, so multi-shot sequences have to be generated separately and stitched together afterward.
How does it keep characters consistent across shots?
LTX-2.5 generates the shots as one connected sequence, so character, environment, lighting, voice, and visual style can carry across cuts. Describing consistent details in the prompt reinforces that continuity.
Can LTX-2.5 be fine-tuned or self-hosted?
Yes. It's open weights, built to be run and fine-tuned on your own infrastructure.
