LTX 2.3 - Extend Video

LTX 2.3 Video Extend takes an existing video and generates new footage that picks up from the end of it.

Upload a clip, choose how many seconds to add, set the mode to extend from the end or beginning, and run. LTX 2.3 reads the motion, style, and content of your source video and generates continuation footage that flows from it. Add a prompt to guide what happens next, or leave it blank and let the model follow the existing motion.

Your clip in. A longer clip out.

How do you use LTX 2.3 to extend a video?

Upload a video, set the duration for how many seconds to add, and choose whether to extend from the end or beginning of the clip. Add a prompt to direct what happens in the new footage. The context slider controls how much of the original clip the model reads before generating.

Input video Upload the clip you want to extend. LTX 2.3 reads the motion, subject movement, and visual style from your source to generate matching continuation footage. Cleaner source video with consistent motion gives cleaner extensions.

Prompt Optional but useful. Describe what should happen in the extended footage: "camera continues panning left", "character keeps walking forward", "smoke drifts upward and dissipates." Without a prompt the model follows the existing motion trajectory. With a prompt you can steer it toward a specific action or ending.

Duration How many seconds of new footage to generate. The default is 5 seconds. Shorter extensions (2 to 4 seconds) tend to stay more coherent with the source. Longer extensions give the model more room to drift. Try shorter durations first and chain multiple runs if you need more length.

Mode Two options: end and beginning.

Use end to generate footage that continues after the final frame of your clip. This is the standard path for lengthening a video forward in time.

Use beginning to generate footage that leads into the first frame of your clip. Useful for adding a lead-in or establishing shot to existing content.

Context Controls how much of your source video the model reads before generating the extension. Higher context values give the model more reference material from your clip. If your video has consistent style or motion you want carried through, increase context. Leave it at the default if your source is short.

What is LTX 2.3 video extend good for?

Lengthening short generated clips, adding lead-ins or endings to existing footage, and building longer sequences by chaining multiple extension passes on the same clip.

The most common use is extending a short text-to-video or image-to-video output that ends too abruptly. Run the clip through LTX 2.3 Extend once or twice to push the length where you need it.

Extending a clip with clear, consistent motion works better than extending one with fast cuts or abrupt scene changes. The model generates continuation based on what it sees in your source, so erratic motion at the end of your clip will produce erratic extension footage.

For long-form video where you need 30 to 60 seconds of coherent output, plan to chain multiple short extensions rather than generating one long pass. Each extension takes the previous output as its new source, and results stay more consistent across shorter generation windows.