Models / LTX / Lip sync

LTX lip sync prompting guide

How to prompt LTX in lip sync / dialogue mode. Every point links its source; community tips are marked. Your agent gets the same guide through the Atlas MCP (get_guide).

Prompt structure

speaker→line quoted with accent→delivery→single take camera→ambience

• Establish the shot (cinematography, scale); set the scene (light, palette, texture, atmosphere); action as a flowing sequence with active verbs; characters with emotion shown physically; camera movement and how subjects look after it; audio. source ↗

What the text should describe

• Dialogue in quotation marks with language/accent; screenplay style (character cues + quoted lines) for multi-beat dialogue. source ↗
• Use a single continuous take when dialogue must stay lip-synced in one framing. source ↗
• Audio-to-video: supply the speech track (audio sets the clip length) plus optional image_uri first frame; prompt supplies the visuals. source ↗

Responds well to

• Best: cinematic single-subject scenes, clear camera language, consistent lighting, described audio; dialogue in quotes with language/accent. source ↗

Duration and limits

• API: generate_audio default true; fps default 24; retake section ≥2 s on inputs ≥73 frames; extend 2–20 s (context+duration ≤505 frames); audio-to-video length follows the audio. source ↗

Multi-shot

• Several shots joined by explicit cuts; re-establish framing, identity and audio at each cut. source ↗
• At every cut say whether music/dialogue/ambience continues or changes, and re-identify subjects. source ↗

Avoid and failure modes

• On-screen text spelling not guaranteed; chaotic physics causes artifacts. source ↗
• Don't paste a Kling/Seedance prompt into LTX unchanged: tag syntax and shot-list formatting don't carry over. Aim for ~4–8 descriptive sentences, present tense. source ↗

Negative prompt

Supported via .

Next

Sources

• docs-dev.ltx.io/open-source-model/usage-guides/prompting-guide.md
• docs.ltx.io/api-documentation/api-reference/async-video-generation/submit-audio-to-video.m
• docs.ltx.io/api-documentation/api-reference/async-video-generation/submit-extend.md
• docs.ltx.io/api-documentation/implementation-guides/prompting-guide.md