LTX 2.3 Retake Video is LTX's video extend & retake model on Versely. This page is its structured prompting reference: the 3 parameters its schema actually exposes, the extend-retake technique that applies to it, copy-ready templates.
Everything here is grounded in the same sources Versely's agent reads — the model's input schema. Where a line is general craft advice rather than a documented fact about LTX 2.3 Retake Video, the page says so.
What LTX 2.3 Retake Video wants
The exact input surface, from the same schema the Versely agent fetches with get_model_input_schema before every generation.
| Parameter | What it does | Values |
|---|---|---|
video_urlreq | Source video (mp4/mov/webm/m4v/gif) | string |
promptreq | Edit instruction (e.g. "Remove the man") — 1-5000 chars | string (min 1, max 5000) |
duration | Duration of segment to retake (2-20) | number (min 2, max 20)default: 5 |
Technique that applies here
Extend/retake: continuing or re-rolling an existing generated clip
- This family covers two different operations, and knowing which one you're on changes what the prompt should say: continuing past the end of a clip (Flux 3 Extend Video, Grok Imagine Extend, VEO 3.1 Extend Video — all take a 'what happens next' prompt) versus redoing a segment inside an existing clip (LTX 2.3 Retake Video — takes an edit instruction, per the schema's own example, 'Remove the man').
- VEO 3.1 Extend Video's duration is locked to exactly '7s' per call — the schema note explains the OpenAPI spec shows a free string but the live endpoint only validates the literal '7s' (a 422 error otherwise, reproduced 2026-07-31); 'up to 30s' describes the total you reach by chaining multiple extend calls, not a single request. Write each extend prompt for one 7-second beat, then issue a fresh extend call — with a fresh prompt — for the next one.
- LTX 2.3 Retake Video's retake_mode enum (replace_audio, replace_video, replace_audio_and_video — default replace_audio_and_video) decides what your edit instruction is allowed to touch. If your prompt is a purely visual change, set retake_mode to replace_video so the original audio track survives untouched — the default silently regenerates both.
Copy-ready templates
Replace the bracketed slots; each template says when it's the right shape.
[WHAT HAPPENS NEXT], camera [dolly in / crane up / steadicam follow / static hold], [LIGHTING CONTINUITY NOTE matching the source clip]
Use when: Extending a VEO 3.1 clip one 7-second beat at a time (duration is locked to '7s' per call).
[Remove/Add/Change] [SUBJECT OR ELEMENT] [from/in] the shot — retake_mode: [replace_video / replace_audio / replace_audio_and_video], start_time: [X]s, duration: [Y]s
Use when: Redoing a specific segment of an existing LTX 2.3 clip rather than continuing past its end (schema's own example: 'Remove the man').
How the Versely agent does this automatically
You can use this page by hand, or let the agent apply the same knowledge. Four real mechanisms — no more, no less:
get_model_input_schema— before generating, the agent looks up LTX 2.3 Retake Video's exact input fields, required fields, allowed values, defaults, and min/max bounds. The parameter table above is that same surface.- The prompt enhancer's family rules — 12 per-family rewrite rules (this model's family isn't one of the 12, so only general enhancement applies) shape how a rough prompt gets rewritten.
- The per-provider speech guide — for TTS scripts, the agent follows a provider-specific tag scheme — not relevant to this model, but it's why voiceover scripts come out marked up correctly.
expand_movie_scene— in movie flows, brief scene ideas are rewritten into detailed cinematic descriptions before generation.
Mistakes that waste generations
- Feeding Grok Imagine Extend an uploaded or external video URL — the schema only accepts a prior Grok Imagine task_id, and external URLs are rejected per the docs note.
- Writing an LTX Retake prompt like a continuation ('then she walks off screen') — the field is an edit instruction over an existing segment defined by start_time and duration, not a description of new footage after the clip ends.
- Leaving LTX Retake's retake_mode on its default (replace_audio_and_video) when you only meant to change the picture — the original audio gets regenerated too unless you explicitly set it to replace_video.
The long-form guide
This page is the structured reference. For the essay treatment — worked examples, failure modes, and narrative — read LTX 2.3 Prompting Guide: Fast Iteration Patterns.
This guide also covers
These siblings share LTX 2.3 Retake Video's prompting-relevant input surface, so their prompting URLs resolve here — tier and pricing differences live on their own model pages:
Frequently asked questions
How long can a LTX 2.3 Retake Video prompt be?+
LTX 2.3 Retake Video's schema caps the prompt at 5,000 characters. Spend them on the subject, motion, and the register its family rewards rather than restating parameters the schema already controls.
Does LTX 2.3 Retake Video support negative prompts?+
No — LTX 2.3 Retake Video's published schema has no negative_prompt parameter. Exclusions have to be phrased positively inside the main prompt, or dropped.
How long can a LTX 2.3 Retake Video generation be?+
Duration is bounded (min 2, max 20). Write one continuous beat sized to that window rather than a multi-act script.
How does the Versely agent know LTX 2.3 Retake Video's parameters?+
Before generating, the agent calls its get_model_input_schema tool, which looks up the exact input fields, required fields, allowed values, defaults, and min/max bounds for the model. Nothing on this page is guessed — it is the same schema surface those tools read.
Does this guide also cover LTX 2 Retake Video?+
Yes. LTX 2 Retake Video share the same prompting-relevant input surface as LTX 2.3 Retake Video, so their prompting URLs redirect here instead of duplicating this page. Tier and pricing differences live on each model's own /models page.
Related prompting guides
Generate with LTX 2.3 Retake Video
LTX 2.3 Retake Video is live in Versely — paste a template above, or just describe what you want and let the agent map it onto the schema for you.