Lightricks LTX 2.3 Reference Video To Video

Reference-video-driven LTX 2.3 generation with optional first and last frame control.

Model Overview

LTX 2.3 Reference Video To Video transforms an input video into a new output guided by a text prompt, while optionally using a start frame and end frame image for stronger framing control. It exposes frame count, size preset, FPS, multiscale generation, and conditioning strength controls.

Best At

• Reinterpreting an existing video with prompt guidance.
• Video-to-video generations that still benefit from first or last frame constraints.
• High-control motion exploration with explicit frame and size settings.

Limitations / Not Good At

• Pricing scales with generated video megapixels, so larger sizes and longer runs get expensive quickly.
• Match Video Length can make the preview estimate rougher if the input clip length is unknown in advance.
• Input footage and prompt direction still need to agree on subject and motion intent.

Ideal Use Cases

• Reworking source footage into a new visual treatment.
• Guided motion restyling while preserving some structural cues.
• Iterating prompt-directed video edits before final compositing.

Input & Output Format

• Input: required prompt and video_url; optional image_url, end_image_url, match_video_length, num_frames, video_size, generate_audio, fps, video_strength, and audio_strength.
• Output: generated video URI returned on response.

Performance Notes

• Cost is tied to output video megapixels, which means both frame count and size preset matter.
• Turning off Match Video Length gives a more predictable estimate because num_frames becomes explicit.