LTX-2

LTX-2 (Lightricks)

Text-and-image-to-video generation with synchronized audio, built as a single unified model.

• Audio + video together. Generates visuals and matching sound (dialogue, music, ambience) in one pass instead of stitching separate models.

• High-resolution output. Designed for native high-res video (up to 4K) with strong temporal consistency across frames.

• Unified diffusion architecture. One foundation model handles motion, scene dynamics, and sound alignment, reducing pipeline complexity.
  Developer-oriented. Designed to respond well to structured prompts, enabling easier testing,

• Open weights. Fully open-source release intended for local runs, customization, fine-tuning, and research use.

• Production-aware design. Built with real creative workflows in mind, supporting structured prompting and extension via LoRAs.

Why pick it for Norman AI?

LTX-2 is a strong choice when you want full audio-visual generation in a single stack. It simplifies experimentation with video + sound, reduces multi-model orchestration, and gives developers direct control over performance, quality, and customization.

response = await norman.invoke(
    {
        "model_name": "ltx-2",
        "inputs": [
            {
                "display_title": "Prompt",
                "data": "A cat playing with a ball on mars"
            }
        ]
    }
)