The AI video model built for production editing
Most AI video models only generate. LTX-2.3 edits. Regenerate specific segments with a new prompt, create smooth transitions between keyframes, and edit audio and video independently. Available via on-prem licensing.
.svg)
Key Capabilities
Edit any moment in a video
Define a start and end time and regenerate only that window using your new prompt. Everything outside it is untouched. Powered by RetakePipeline via /v1/retake.
Keyframe interpolation: Generate transitions between shots
Give LTX two reference frames and a prompt. It generates a smooth, AI-driven transition between them, connecting shots, filling gaps, or turning storyboard frames into moving sequences.
Edit audio and video independently
Edit audio and video streams separately within any time window. Regenerate the video while keeping the original audio. Regenerate the audio–voiceover, score, or SFX–while keeping every frame
Fix a specific moment. Keep everything else.
A finished clip has one weak scene. Define the window. Write a corrected prompt. Regenerate only that segment. The rest of the clip is exactly as you left it. You're paying for the seconds you're changing, not re-rendering the full clip from scratch.
Generate transitions between shots with keyframe interpolation
Two reference images: the end of one shot and the start of the next. LTX-2.3 generates the visual transition between them, cinematic cuts, scene interpolation for pre-vis, or storyboard frames turned into moving sequences. Powered by KeyframeInterpolationPipeline. No timeline software required.
.jpeg)
Replace audio without touching the video
Set regenerate_audio to True and regenerate_video to False. Within the defined window, the audio track is regenerated and the video frames are untouched. Use it for localization, temp music swaps, or iterating on a voiceover without redoing the cut.
Automated editing and QA at scale
In automated generation pipelines, use /v1/retake to detect and replace problem segments without discarding the full output. Catch a weak moment, trigger a retake, swap only that segment, continue. No manual review cycles.
Designed for real-world deployment
Production-ready AI video editing model for enterprise teams building scalable, controllable video generation workflows.
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 LTX-2.3’s Retake video editing works
Input
Technical characteristics:
Segment Retake (/v1/retake):
- Source video file
- Frame count must satisfy 8k+1 format (e.g. 97, 193, 257, 385 frames)
- Resolution must be multiples of 32 (e.g. 1920x1080, 1280x720)
- Start_time and end_time (defines the editing window in seconds)
- Text prompt for the regenerated segment
- Regenerate_video: True or False
- Regenerate_audio: True or False
- Model: ltx-2-fast or ltx-2-pro
Keyframe Interpolation (KeyframeInterpolationPipeline):
- Two keyframe images (start frame and end frame)
- Text prompt guiding the transition
- Target duration and resolution
Output
Technical characteristics:
Segment Retake:
- Full video returned with the specified segment replaced; content outside the window is bit-for-bit identical to source
- Regenerated segment is temporally coherent with surrounding content
Keyframe Interpolation:
- Generated transition video between the two input keyframes
- Same resolution as input keyframes
Both pipelines:
- Available via LTX API (REST) or run open-source on 80GB+ VRAM GPU (32GB with FP8 quantization)
Retake - Video Editing
Refine only the parts that need adjustment - no need to regenerate the whole video. Perfect for fixing scenes, adjusting elements, or improving localized areas.
- Currently available in 1080p only.
- Billed per second of input video.
Retake - Video Editing
Refine only the parts that need adjustment - no need to regenerate the whole video. Perfect for fixing scenes, adjusting elements, or improving localized areas.
- Currently available in 1080p only.
- Billed per second of input video.
FAQs
What kinds of video edits can LTX-2.3 make?
LTX-2.3 supports three editing modes: temporal segment regeneration (replace any time window with a new prompt via Retake), keyframe-guided transition generation (KeyframeInterpolationPipeline), and independent audio/video stream editing, video only, audio only, or both within a defined window.
Can I edit a specific part of a video without re-rendering the whole clip?
Yes. /v1/retake lets you define a start_time and end_time. Only that segment is regenerated, everything outside it is bit-for-bit identical to the source. You pay per second of the segment being edited.
Can I replace audio without changing the video?
Yes. Set regenerate_audio to True and regenerate_video to False. The video frames within the window stay exactly as they are. Only the audio stream is regenerated.
What is the 8k+1 frame count requirement?
The source video's total frame count must equal 8k+1 for some integer k (e.g. 97, 193, 257, 385 frames). This is a constraint of RetakePipeline. Trim or pad the source to the nearest valid count before calling /v1/retake. Resolution must also be in multiples of 32.
