LTX-2.5
LTX's 22B open model that makes video and synchronized audio together, with native multi-shot scenes; free below US$10M revenue.
In short. LTX-2.5 is an AI video model you can download and run yourself. From a text prompt, an image or a clip it produces video with matching sound, and it can create several connected shots in one go with the same characters and voices. It is free for individuals and companies earning under US$10 million a year; bigger companies need a paid license.
Summary
LTX released LTX-2.5 on 11 August 2026 as open weights. It has 22 billion parameters and, according to LTX, is an asymmetric dual-stream diffusion transformer with cross-attention between a video stream and an audio stream, so picture and sound are generated together in a single pass. Inputs can be text, images, audio or video: text-to-video, image-to-video, audio-to-video, first-and-last-frame interpolation and video-to-video are all supported.
LTX lists these changes from LTX-2.3: native multishot generation, which makes connected shots that keep character identity, setting, lighting and voice across cuts; a diffusion video decoder that replaces the older VAE stage for sharper faces, text and motion; a custom Gemma 4 12B text encoder; an optional prompt enhancer and an optional duration predictor; and an improved distilled model with a fixed 8-step schedule and no guidance. ComfyUI documents native 4K HDR output at up to 50 frames per second.
The weights are split into components: a 22B transformer (about 42 GB in BF16, available as dev and distilled checkpoints), the Gemma 4 text encoder (about 26 GB), video and audio autoencoders and upscalers, which adds up to roughly 71 GB for the full BF16 set. ComfyUI-only INT8 builds bring the distilled set to about 40 GB, and a distilled NVFP4 transformer of about 19 GB is aimed at Blackwell GPUs. LTX's own AI-information page says the full model needs a GPU with 80 GB or more, while launch coverage quotes LTX's claim that optimized builds run on GPUs with at least 16 GB. LTX also publishes many task-specific LoRAs on Hugging Face, for example for restoration, colorization, deblurring and SDR-to-HDR conversion.
LTX was split off from Lightricks as its own business unit in June 2026. The weights are under the LTX-2.x Community License: free commercial use for organizations with annual revenue under US$10 million (counted together with affiliates) and a paid agreement above that. The license also requires you to keep the model's watermarking and content-provenance features and to follow AI transparency laws such as the EU AI Act. The download is gated on Hugging Face, and LTX's speed and preference-test results are its own and have not been independently confirmed.
Good for
- Private, local video generation with synchronized sound, without per-clip fees
- Short multi-shot scenes in which characters and voices stay consistent across cuts
- Fine-tuning and LoRA work, supported by LTX's trainer and its library of task LoRAs
- Individuals, studios and companies under US$10 million in revenue
Watch out for
- Companies at US$10 million revenue or more (counted with affiliates) need a paid commercial license
- You must keep watermark and provenance features and meet AI disclosure laws; removing them can end your license
- The full BF16 set is about 71 GB and LTX says it needs 80 GB or more of GPU memory; the 16 GB route uses optimized builds
- It is not Apache 2.0 or open source in the strict sense, and the Hugging Face download is gated
- Speed and quality claims come from LTX and have not been independently confirmed
Performance claims come from the developer and have not been independently verified.
How much memory does it need?
Memory figures for each variant of LTX-2.5. Image, video, speech and music pipelines also load a text encoder and a decoder, so the memory needed can be larger than the main checkpoint.
| Variants | Parameters | Download size | Memory needed | Fits on |
|---|---|---|---|---|
| Dev and Distilled (BF16) BF16 | 22B | 71 GB | 80 GB | Workstation |
| Distilled INT8 (ComfyUI) INT8 | 22B | 40 GB | 16 GB | Laptop or phone |
Download size covers every component needed to run the model, such as the generator, text encoder and decoder. Memory needed is a stated or published measured minimum where one exists; otherwise it is the download size plus about 15%. Larger images, longer clips and higher resolutions need more. The developer states a minimum of 16 GB of GPU memory.
The license in plain words
A custom license for LTX video models. Organizations with annual revenue under US$10 million can use, modify and host the model for any purpose; organizations at or above that figure need a paid commercial license, though they may still evaluate it in non-production settings.
- Free commercial use only below a revenue cap
- Outputs must be filtered or reviewed, and AI generation disclosed
- Acceptable-use policy applies
See also
- LTX-2.5 vs Wan 2.2: open-source video generation compared
- Best open-source image and video generation models
- Cosmos 3: NVIDIA's open world models for robots and driving
- Free until you grow: where open licenses draw the line
Frequently asked questions
- What is LTX-2.5?
- LTX-2.5 is an AI video model you can download and run yourself. From a text prompt, an image or a clip it produces video with matching sound, and it can create several connected shots in one go with the same characters and voices. It is free for individuals and companies earning under US$10 million a year; bigger companies need a paid license.
- Can I use LTX-2.5 commercially?
- Yes, with conditions. LTX-2.5 is released under the LTX-2.x Community, which allows commercial use but adds extra rules: Free commercial use only below a revenue cap; Outputs must be filtered or reviewed, and AI generation disclosed; Acceptable-use policy applies. Read the license text before you ship a product.
- Is LTX-2.5 open source?
- LTX-2.5 is open-weight, not open source in the strict sense. You can download the model files, but the LTX-2.x Community adds conditions that the Open Source Initiative definition does not allow.
- Can I run LTX-2.5 on my own computer?
- Yes. The smallest variant, Distilled INT8 (ComfyUI), needs about 16 GB of memory at 4-bit precision, which fits a modern laptop or a mid-range graphics card.
- When was LTX-2.5 released?
- LTX-2.5 was released by LTX (Lightricks) on August 11, 2026.
Official links
Sources
Specifications come from these primary sources. Benchmark claims, where mentioned, are reported by the developer.
- LTX-2.5 model card (Hugging Face) huggingface.co
- LTX: official AI information page ltx.io
- LTX-2.5 product page ltx.io
- VentureBeat: LTX-2.5 launch coverage venturebeat.com
- ComfyUI docs: LTX-2.5 workflows docs.comfy.org
- LTX-2.x Community License Agreement github.com
Release date and facts last verified November 8, 2026.
Related models
| Model | Released | Parameters | Context | License | Smallest variant needs |
|---|---|---|---|---|---|
| Wan 2.2 Alibaba Wan | 5B – 27B 14B active | – | Apache 2.0 (Permissive) | 24 GB Gaming GPU | |
| FLUX 3 Image Klein Black Forest Labs | 5.6B | – | FLUX Kommunity (Conditional) | 8 GB Laptop or phone | |
| MiniMax H3 MiniMax | 33.1B | – | MiniMax H3 Community (Conditional) | 166 GB Big-memory desktop | |
| Pika H3 Flex Pika | 33.1B | – | MiniMax H3 Community (Conditional) | 169 GB Big-memory desktop |