Video Editing

LTX-2.3 provides several documented ways to modify an existing video with generative AI. The most direct editing operation is Retake, which regenerates a selected section of an existing video from a text prompt while preserving the rest of the video. LTX-2.3 Pro also supports Extend, which generates additional frames at the beginning or end of a video, and Reframe, which changes the composition to a new resolution by generating content in areas that were not present in the original frame. These capabilities are designed for iterative video generation and modification rather than traditional cuts, transitions, or timeline-based editing.

The LTX-2.3 Pro Retake endpoint regenerates a defined section of an existing video using a text prompt. The section is specified with a start time and duration. The operation can replace the video, the audio, or both audio and video, while the rest of the source video is preserved. This makes Retake suitable for iterative correction when a particular section of a generated clip needs another result without regenerating the complete video.

In the open-source LTX pipeline implementation, Retake encodes the source video and audio into latent representations and applies a temporal mask to the selected interval. Only the masked region is regenerated, while content outside the selected time window is preserved. The implementation also supports independently regenerating video and audio. This gives Retake a more precise scope than simply generating a new clip from scratch.

The LTX-2.3 Pro Extend endpoint generates additional frames at either the beginning or end of an existing video. The model uses context frames from the input video to generate a continuation with consistent motion and audio. The API accepts a prompt describing what should happen in the extended portion and supports an extension duration from 2 to 20 seconds. This makes Extend useful when a generated shot needs to continue beyond its original endpoint without creating a completely separate clip.

LTX also documents a Reframe operation for changing an existing video's composition to a new resolution. Instead of merely cropping the source, the operation fills areas that are missing in the new composition with generated content intended to match the original video. This is particularly relevant when the same source material needs to be adapted to a different frame shape or resolution. Reframe is therefore a generative composition operation, not a conventional resize filter.

Video-to-Video (V2V) uses an existing video as conditioning for a transformation, with the open LTX ecosystem providing IC-LoRA-based V2V workflows. By contrast, the documented LTX-2.3 editing operations work directly on an existing generated video for a defined purpose: Retake regenerates a selected temporal region, Extend adds new frames before or after the clip, and Reframe changes the composition by generating missing areas. Keeping these workflows separate makes it easier to choose the appropriate LTX capability for a specific task.

The documented capabilities focus on specific generative operations. Retake handles regeneration of a selected section, Extend handles temporal continuation, and Reframe handles generative adaptation to a new composition. They do not constitute a general-purpose timeline editor with conventional cutting, transitions, keyframing, color grading, or compositing tools. For workflows that require those operations, LTX can be used as part of a broader production pipeline rather than as the entire editing environment.

The official LTX model documentation lists Retake, Extend, and Reframe as supported by ltx-2-3-pro. They are not available through the LTX-2.3 Fast API model. The API therefore separates rapid T2V/I2V generation from the more advanced editing-oriented operations available through Pro. This distinction is important when planning an API-based workflow around LTX-2.3.

Use Retake when a specific section of an existing video needs to be regenerated. Use Extend when the shot should continue beyond its current beginning or end. Use Reframe when the composition needs to be adapted to a new resolution and the missing image areas should be generated. If the goal is instead to transform an existing video according to reference-video conditioning, the appropriate concept is V2V, particularly the IC-LoRA-based workflows in the open LTX ecosystem.