LTX-2.3.
LTX-2.3 is an open-weight video model from Lightricks. Specs, recommended settings, hardware and licence terms, from the developer's own pages.
- Status
- Reference
- Task
- Text to video, Image to video
- Input
- Prompt, optional images, video or audio
- Output
- Video with synchronised audio
- Developer
- Lightricks
- Parameters
- 22B
- Licence
- LTX-2 Community License
- Released
- March 2026
What the model does.
Paraphrased from the official model card.
LTX-2.3 is an update to Lightricks' LTX-2, a DiT-based model that generates video and synchronised audio in one model and is built for running locally. Its card describes improved audio, visuals and prompt adherence over LTX-2.
It's published as single-file 22B checkpoints that bundle the transformer, VAEs and text projection: a trainable dev model, distilled models that run 8 steps at CFG 1, a distilled LoRA, and spatial and temporal upscalers. The Gemma 3 text encoder is downloaded separately. Lightricks now lists it as legacy next to LTX-2.5, but its pipelines still run on it.
Reference page.
We don't currently serve LTX-2.3. The specs, settings and licence terms on this page come from Lightricks's own pages.
Is LTX-2.3 licensed for commercial use?
Licence: LTX-2 Community License.
Conditional. Organisations with less than $10 million in annual revenue, counted with their affiliates, can use it commercially for free, including hosting it as a service. At $10 million or more you need a paid commercial licence from Lightricks. The licence also bars using it in a product that directly competes with Lightricks' own commercial products without a separate licence.
- LTX-2 Community License, dated January 5, 2026, covering LTX-2 versions released since then. Free for any purpose under the Attachment A restrictions, but entities with annual revenue of at least $10,000,000 need a paid commercial use licence. Revenue counts all subsidiaries and affiliates under common control. LTX-2 repo, LICENSE-2
- Section 2: unlicensed commercial use by such an entity is a material breach, owing licence fees plus liquidated damages of double the amount that would have been paid. LTX-2 repo, LICENSE-2
- Section 3: hosting for third parties (for example as software as a service) is allowed if Attachment A is an enforceable part of your user terms, users are told about it, and recipients get a copy of the agreement. Derivatives, including fine-tuned weights, must be distributed under this licence. LTX-2 repo, LICENSE-2
- Attachment A, item 20: no use in any product, service or application that directly competes with Lightricks' commercial products or services, or is designed to replace them, without a separate commercial licence. Item 18: for commercial use, no training, improving or fine-tuning other models with it. LTX-2 repo, LICENSE-2
- Attachment A also requires disclosing machine-generated content, and bars deepfakes without consent, medical advice, military use and fully automated decisions affecting legal rights. Lightricks claims no rights in outputs. New York law, with ICC arbitration in New York. LTX-2 repo, LICENSE-2
This is our reading of the licence, not legal advice. Read the licence text before you rely on it.
LTX-2.3 hardware requirements.
Published by Lightricks for running the weights yourself. These are the developer's figures, not Windpaint's.
- Published VRAM
- Not stated
- Applies to
- No inference VRAM figure. Tested with Python 3.12+, CUDA above 12.7; FP8 casting and offload suggested for limited memory.
- Neither the card nor the README gives a VRAM figure for inference. For GPU memory constraints the README suggests
--quantization fp8-cast --offload {cpu, disk}; fp8-cast downcasts the bf16 checkpoint on the fly. LTX-2 repo - The card says the codebase was tested with Python 3.12 or newer and CUDA above 12.7, and supports PyTorch ~2.7. LTX-2.3 model card
- Each checkpoint is a single file bundling the transformer, the video and audio VAEs and the text projection. The Gemma 3 text encoder is downloaded separately. LTX-2 repo, MODELS-LTX-2.3.md
- The LTX-2 README now lists LTX-2.3 as legacy, with LTX-2.5 recommended, but every pipeline in the repo still runs on it. Files aren't interchangeable between the two. LTX-2 repo
- The LTX-2 Trainer (LoRA, full fine-tuning and IC-LoRA) recommends Linux with CUDA and an NVIDIA GPU with 80GB+ VRAM for its standard config. Its low-VRAM config, with INT8 quantization, targets 32GB cards such as the RTX 5090. LTX-2 repo, README.md
LTX-2.3 native resolution.
What the developer's own code and examples use.
- Width and height must be divisible by 32, and the frame count must be divisible by 8, plus 1. If they aren't, pad the input with -1 and crop to the size and frame count you want. LTX-2.3 model card
- Upscalers: x2 and x1.5 spatial upscalers for multi-stage pipelines at higher resolution, and an x2 temporal upscaler for higher frame rates. LTX-2.3 model card
Recommended settings for LTX-2.3.
Steps, guidance and defaults from the developer's card and code, for running the weights yourself.
ltx-2.3-22b-devis the full model, trainable in bf16.ltx-2.3-22b-distilledandltx-2.3-22b-distilled-1.1run 8 steps at CFG 1; the card says 1.1 has a different aesthetic and improved audio compared with 1.0. LTX-2.3 model card- The distilled LoRA (rank 384) applies the distilled model to the full model. LTX-2.3 model card
- The two-stage pipelines need the spatial upscaler, and all but DistilledPipeline, ICLoraPipeline and DubItPipeline also need the distilled LoRA. LTX-2 repo, MODELS-LTX-2.3.md
- For ComfyUI the card recommends the built-in LTXVideo nodes from the ComfyUI Manager. LTX-2.3 model card
LTX-2.3 prompt guide.
From Lightricks's own prompt guide, prompt tooling and examples.
- The LTX-2 README: write detailed, chronological descriptions of actions and scenes in a single flowing paragraph, start directly with the action, keep descriptions literal and precise, and stay within 200 words. LTX-2 repo
- Its suggested structure: the main action in one sentence, then movements and gestures, appearances, background and environment, camera angles and movements, lighting and colours, and any changes or sudden events. LTX-2 repo
- The README's example prompt writes sound into the description: a quick sniff is heard, then the man speaks "with a deep male voice and a satisfied tone" and his lines are quoted. LTX-2 repo
- The card points to Lightricks' prompting guide for LTX-2 for more tips. LTX-2.3 model card
LTX-2.3 limitations.
As the developer states them.
- Not intended or able to provide factual information, and may amplify existing societal biases. LTX-2.3 model card
- Prompt following is heavily influenced by prompting style, and videos may not match the prompt perfectly. LTX-2.3 model card
- May generate content that is inappropriate or offensive. LTX-2.3 model card
- When generating audio without speech, the audio may be of lower quality. LTX-2.3 model card
Questions.
Can I call LTX-2.3 on Windpaint?
Not at the moment.
How much VRAM does LTX-2.3 need?
Lightricks doesn't give a VRAM figure for inference. For limited GPU memory the LTX-2 README suggests FP8 casting with CPU or disk offload.
Can I use LTX-2.3 commercially?
Yes if your organisation, counted with its affiliates, has less than $10 million in annual revenue, and hosting it as a service is allowed. At $10 million or more you need a paid commercial licence from Lightricks. Use in a product that directly competes with Lightricks' own needs a separate licence.
Is LTX-2.3 open source?
No, it's open-weight. The weights are on Hugging Face under the LTX-2 Community License Agreement. That's Lightricks' own licence, with a $10 million revenue threshold and use restrictions, not a standard open-source one.
How do I train a LoRA for LTX-2.3?
Use the LTX-2 Trainer, which supports LoRA, full fine-tuning and IC-LoRA on LTX-2.3's dev model. It recommends an 80GB+ GPU for the standard config and has a low-VRAM INT8 config for 32GB cards such as the RTX 5090.
Is LTX-2.3 still supported?
Lightricks' README lists it as legacy and recommends LTX-2.5, but every pipeline in the LTX-2 repo still runs on LTX-2.3. The two models' files aren't interchangeable.
Does LTX-2.3 work in ComfyUI?
Yes. The card recommends the built-in LTXVideo nodes, which you can find in the ComfyUI Manager.
Is there an LTX-2.3 prompting guide?
Lightricks suggests one flowing paragraph under 200 words, in chronological order: the main action, then movement, appearance, environment, camera, lighting and any sudden changes. The card also links Lightricks' prompting guide.