Source: https://ltx.io/model/model-blog/ltx-2-3-prompt-guide

If this page helps, consider buying me a coffee 🍵

Prince Ecuacion

Overview

LTX-2.3 responds best to detailed prompts that describe the full scene clearly, including subject, action, environment, lighting, camera movement, and audio. It is more faithful to prompt details than earlier versions, so specificity usually improves results.

Key Takeaways

• Use long, descriptive prompts instead of short generic ones.
• Describe the full scene, not just the subject.
• Use cinematic language like shot type, lens feel, movement, and lighting.
• Include audio when generating synchronized audio-video.
• Break spoken dialogue into shorter phrases with acting beats between them.
• Match prompt length to video length; longer videos need more direction.
• Use physical cues instead of abstract emotional labels.
• For image-to-video, focus more on what moves next than on what is already visible.

Core Principles

1. Be Specific and Descriptive

Instead of writing:

• a person walking

Write something more like:

• a young woman in a red coat walking briskly through a rain-soaked Tokyo street at night, neon reflections on wet pavement, handheld camera following from behind