LTX 2.5 AI Video Generator

LTX 2.5 is Lightricks' open-weight audio-video foundation model, released on August 11, 2026. It uses a 22B asymmetric dual-stream diffusion transformer to generate video and synchronized audio, and adds native multishot continuity, a diffusion video decoder, automatic duration, stronger prompt understanding, Diffusion Fidelity Rendering, and professional HDR/RAW workflows.

LTX 2.5 supports text-to-video, image-to-video, native multishot sequences, audio-conditioned generation, keyframe interpolation, retake, extension, and adapter-driven transformations. Available controls depend on the selected workflow, but prompts should describe the action, camera, lighting, sound, and shot transitions in chronological order.

The largest changes are native connected multishot generation, a new diffusion video decoder, a custom Gemma 4 12B text encoder, a dedicated prompt enhancer, prompt-predicted duration, Diffusion Fidelity Rendering, native EXR/HDR/RAW support, a stronger distilled model, and a raw pretrained checkpoint for deeper adaptation. Many LTX 2.3 LoRAs and IC-LoRAs may work, but Lightricks recommends validating adapters because compatibility is not universal.

Yes. The official Fast and Pro tiers list 1080p, 1440p, and 4K output at 24, 25, 48, or 50 fps. The native HDR workflow can accept EXR inputs in supported scene-linear or ACES-family color spaces and write half-float EXR frames plus a 10-bit BT.2020/HLG HEVC master. Fast supports up to 20 seconds at 1080p and up to 10 seconds at 1440p or 4K; Pro supports up to 10 seconds.

Yes. LTX is a joint audio-video model with separate video and audio streams connected through bidirectional cross-modal attention. Supported pipelines can generate sound with the picture and expose modality-specific guidance. Prompt dialogue, delivery, ambience, effects, and music as part of the same chronological scene. As with any generative model, multi-speaker identity and long sequences should still be reviewed before production use.

The weights and inference/training code are publicly available under the LTX-2.x Community License. Lightricks states that entities below $10 million in total annual revenue can use it commercially and in production at no cost, subject to the complete license terms; larger entities need a paid commercial agreement. Fine-tune transfer and redistribution can carry additional conditions, so the license—not a landing-page summary—should govern deployment decisions.

The recommended split distilled download is roughly 66 GiB before optional components. The official Python stack recommends Python 3.12 or newer, CUDA 12.7 or newer, and a modern PyTorch environment. FP8 casting, NVFP4 on Blackwell GPUs, CPU or disk offload, tiling, and different attention backends can reduce memory pressure, but high-resolution audio-video generation remains a serious GPU workload rather than a lightweight desktop task.

LTX's launch page lists official API pricing per generated second: $0.09 at 720p, $0.15 at 1080p, $0.19 at 2K, and $0.37 at 4K. Those are LTX's published API prices rather than Musci3 credit prices. The applicable Musci3 credit consumption is displayed in the generator before submission.