LTX 2.3 - Retake Video

LTX 2.3 Retake lets you fix or replace a specific section of an existing video without regenerating the whole clip.

Upload a video, write a prompt for what the segment should look like, set where in the clip to start and how many seconds to cover, then choose whether to replace the audio, video, or both. LTX 2.3 regenerates that window and stitches the result back in.

No need to rerun the entire clip to fix one bad moment.

How do you use LTX 2.3 retake?

Upload your video, write a prompt describing the replacement segment, set the start time and duration, and pick a retake mode. LTX 2.3 regenerates that window of your clip according to your prompt and replaces the original footage.

Input video Upload the clip you want to fix. LTX 2.3 reads the surrounding context to generate a replacement segment that fits the visual style and motion of your original. Clips with consistent lighting and movement across the retake window blend more naturally.

Prompt Describe what the regenerated segment should look like. Be specific about the action, camera angle, and subject behavior. "The man walks forward on the street and looks around casually with subtle body motion." That level of specificity gives the model a clear target for the replacement segment.

Start time The timestamp in seconds where the retake begins. Set this to the frame where the problem starts. The model uses the footage before this point as context for generating the replacement.

Duration How many seconds to regenerate. Default is 5 seconds. Keep this tight around the section you want to fix. A narrower retake window stays more coherent with the surrounding footage on both sides.

Retake mode Three options that control what gets replaced:

replace_audio_and_video: regenerates both the visuals and audio for the selected segment. Use this when the whole segment needs replacing.

replace_video_only: regenerates the visuals while leaving the original audio intact. Use this when timing or dialogue in the audio is correct but the motion or appearance needs a fix.

replace_audio_only: regenerates only the audio while keeping the original visuals. Use this when the footage is fine but the sound needs work.

What is LTX 2.3 retake good for?

Fixing specific moments in generated video without rerunning the full clip. Correcting motion artifacts, awkward transitions, bad frames, or misaligned audio in a targeted segment.

The most direct use is cleanup: a clip came out well except for one 3 to 4 second window where the motion glitched or the subject looked wrong. Point retake at that segment, write a prompt describing the correct version, and regenerate only what needs fixing.

Retake also works for targeted reshoots. If a specific action in your clip needs to read differently: a gesture that's too fast, a look that doesn't land, a camera move that breaks the flow. You can replace that moment without touching the rest of the clip.

For changes that span the full length of the video, or structural edits that affect the whole composition, a full vid2vid workflow will give you better results. Retake is for surgical fixes on specific windows, not whole-clip transformations.