ltx/ltx-2/retake-video

What makes LTX 2 Video Retake stand out

LTX 2 video retake is a high-fidelity video-to-video model built for precise retakes that preserve spatial layout, motion continuity, and scene identity. Given a source clip, LTX 2 performs targeted transformations rather than full-frame regeneration, maintaining believable structure while updating style, content, or pacing. With unified audio-video generation, LTX 2 can replace visuals, audio, or both in sync, minimizing drift across edits. Support for 4K motion and fast modes allows LTX 2 to deliver cinematic detail with production-friendly turnaround times. Granular control over start time and duration helps LTX 2 operate surgically on segments, keeping the surrounding footage intact for seamless joins. Open-source access enables teams to adapt LTX 2 to varied pipelines while retaining consistent realism and temporal stability. In complex scenes, LTX 2 balances adherence to the base camera motion with prompt-driven changes. Key capabilities:

• Structure-preserving edits that maintain pose, layout, depth, and motion cues to avoid warping or re-synthesis artifacts.
• Unified audio-video retakes: replace audio, replace video, or both with tight temporal synchronization.
• Segment control via start_time and duration for localized changes while protecting the rest of the timeline.
• Realistic continuity across frames with stable lighting, shadows, reflections, and minimized flicker.
• Style and content restyling for cinematic grades, texture updates, and object swaps without breaking composition.
• Production-ready performance with 4K motion support and fast modes for iterative turnarounds.

Prompting guide for LTX 2

Begin by supplying a clear video_url and a concise prompt describing the intended retake; specify what to change and what to preserve so LTX 2 video retake targets the correct region. Use start_time and duration to define the exact segment, allowing LTX 2 to work nondestructively within in-out boundaries. Choose retake_mode to replace_audio, replace_video, or replace_audio_and_video; LTX 2 will synchronize timing and maintain alignment with the source. For visual edits, describe motion, style, and camera cues; for audio edits, provide timing or dialogue intent so LTX 2 video retake can align delivery and ambience. Iterate with concise refinements and, when needed, provide references to guide color, texture, or mood while LTX 2 preserves the original scene geometry. When upscaling or finishing, keep shot cadence and edit points consistent so LTX 2 video retake can render transitions cleanly. Example prompts and cases:

• Replace_video, 0-5 s: preserve subject and camera path; restyle background to overcast city, subtle cinematic grade.
• Replace_audio, 2-7 s: keep all visuals; swap ambient crowd sound with quiet interior tone, maintain natural reverberation.
• Replace_audio_and_video, 5-10 s: convert midday to golden hour, add soft wind ambience; preserve actor pose and pacing.
• Localized background change only: keep skin tones, wardrobe, and lens bokeh; replace skyline with foggy harbor.
• Object removal: remove a moving boom mic in the top-right; maintain reflections and consistent grain. Pro tips:
• State hard constraints first so LTX 2 prioritizes them.
• Use precise spatial and temporal language like background only, lower third, 0-3 s region.
• Avoid conflicting adjectives; prefer a few strong descriptors that match the footage.
• Feed clean, stable sources; trim or stabilize shaky handles before processing.
• Iterate with short updates, compare versions side by side, and lock what works before expanding scope.