LTX-2.5-22b-IC-LoRA-Restore Video Generation Model: File List

October 1, 2026

LTX-2.5-22b-IC-LoRA-Restore Video Generation Model: File List

At a Glance

Item Value
Repository Lightricks/LTX-2.5-22b-IC-LoRA-Restore
Publisher guide Lightricks (LTX): models and licenses
Published 2026-09-28
License ltx-2.x-community-license
Formats safetensors
Access Gated on Hugging Face (license agreement required)
Source type Primary source (the publisher itself)

Values determined by this site’s code when the information was collected. Dates are JST.

Overview

Lightricks has released “Lightricks/LTX-2.5-22b-IC-LoRA-Restore", a video restoration adapter model. This model is an IC-LoRA model supporting “video-to-video" tasks where input videos are processed to generate new videos, using the company’s audio-video foundational model “Lightricks/LTX-2.3" as its base.

According to tag information, the model is built for applications such as archiving video restoration, black-and-white video colorization, and VFX processing. Accessing the model requires agreeing to terms of use on Hugging Face.

Specifications

Specifications listed in the model card and metadata are as follows:

  • Architecture: Video-to-video IC-LoRA adapter (the base model Lightricks/LTX-2.3 is a DiT-based diffusion foundation model)
  • Base Model: Lightricks/LTX-2.3
  • Output Specifications and Resolution Constraints (recommended specs of the base model): The width and height of generated video must be divisible by 32. In addition, the frame count must be a multiple of 8 plus 1. If the specified resolution or frame count does not meet these conditions, the documentation notes that input data should be padded with “-1" and cropped to the target resolution and frame count after generation.
  • Supported Languages: English (en)
  • Distribution Format: safetensors format
  • License Terms: The “ltx-2.x-community-license" applies to this model (the base model uses ltx-2-community-license-agreement). It is a gated model requiring agreement to the license upon acquisition.

The list of checkpoints published in the repository for the base model “LTX-2.3" is as follows:

Name Notes
ltx-2.3-22b-dev The full model, flexible and trainable in bf16
ltx-2.3-22b-distilled The distilled version of the full model, 8 steps, CFG=1
ltx-2.3-22b-distilled-1.1 The distilled v1.1 version of the full model, 8 steps, CFG=1 – A different aesthetic experience and improved audio compared to v1.0
ltx-2.3-22b-distilled-lora-384 A LoRA version of the distilled model applicable to the full model
ltx-2.3-22b-distilled-lora-384-1.1 A LoRA version of the v1.1 distilled model applicable to the full model
ltx-2.3-spatial-upscaler-x2-1.1 An x2 spatial upscaler for the ltx-2.3 latents, used in multi stage (multiscale) pipelines for higher resolution
ltx-2.3-spatial-upscaler-x1.5-1.0 An x1.5 spatial upscaler for the ltx-2.3 latents, used in multi stage (multiscale) pipelines for higher resolution
ltx-2.3-temporal-upscaler-x2-1.0 An x2 temporal upscaler for the ltx-2.3 latents, used in multi stage (multiscale) pipelines for higher FPS

Performance and Quality

Specific benchmark measurements for this LoRA model alone are not provided in the materials, but official guidance regarding the performance and quality characteristics of the base “LTX-2.3" model is as follows:

It is described that the base model LTX-2.3 was developed as a major update to its predecessor LTX-2, featuring improved video and audio quality as well as enhanced prompt adherence. It is designed to generate video-synchronized audio within a single model, focusing on practical execution in local environments as an open-weight model.

On the other hand, the official documentation clearly outlines constraints and cautions regarding the base model. Points include that the model is not intended to provide factually accurate information, and that it may fail to generate videos that completely match the prompt. It is also indicated that prompt adherence strongly depends on the description style of the input prompt, and that audio synthesis quality may degrade when generating sounds that do not include speech.

Strengths and Use Cases

“Lightricks/LTX-2.5-22b-IC-LoRA-Restore" is an IC-LoRA model specialized for “video-to-video" tasks that take existing video as input and convert or correct it into new video output.

Target use cases and strengths are clearly indicated by the tag information attached to the model card. Specifically, they include video restoration of archival footage and historical materials, colorization of monochrome/black-and-white video material, and application of various visual effects (VFX) or correction processing in video production. It is well-suited for scenarios where quality improvement or coloring is desired while maintaining the structural and motion context of the original video frames.

The base model “Lightricks/LTX-2.3" is a DiT-based foundation model capable of simultaneously generating video and synchronized audio within a single model. The base model itself supports a wide range of tasks including text-to-video, image-to-video, and cross-modal audio-video translation, featuring multilingual environments and expressive generation capabilities.

Furthermore, the design of the base model emphasizes flexible fine-tuning and training in local environments. According to official documentation, training for behaviors, styles, or representations of appearance and sound can be completed in under an hour in many settings using the official training tool (ltx-trainer). This model is constructed leveraging such high extensibility and tuning performance, acting as a control adapter specialized for video restoration and specific effects.

Distributed Files

Weight files published in Lightricks/LTX-2.5-22b-IC-LoRA-Restore, listed by this site from the Hugging Face API. Sizes are the actual file sizes.

File Size
ltx-2.5-22b-ic-lora-restore-1.0.safetensors 1.71GB

How to Get It

Distribution formats, download notes, and recommended execution environments for this model are as follows:

  • Distribution Format and Access Procedures:

  • Model weight files are released in safetensors format. – This model (Lightricks/LTX-2.5-22b-IC-LoRA-Restore) is managed as a gated model on Hugging Face. Therefore, to actually download and use the files, you must access the Hugging Face model page ( https://huggingface.co/Lightricks/LTX-2.5-22b-IC-LoRA-Restore ) and agree to the presented terms of use and license conditions (ltx-2.x-community-license).

  • Supported Tools and Execution Frameworks:

  • When running in ComfyUI, official documentation recommends using the built-in “LTXVideo" nodes available via ComfyUI Manager. – The official PyTorch codebase is published in the GitHub “LTX-2" repository ( https://github.com/Lightricks/LTX-2.git ). The repository uses a monorepo structure, and the ltx-pipelines package is used to execute inference pipelines. Verified environments by the developers include Python 3.12 or higher, CUDA versions above 12.7, and an operating environment equivalent to PyTorch 2.7. – Regarding direct support in the popular Diffusers library, official documentation states it is “coming soon".

Can You Run It Locally?

The publisher distributes this model as safetensors.

License — ltx-2.x-community-license (Commercial use allowed with conditions): Lightricks’ LTX-2.x Community License. Businesses with annual revenue of $10 million or more need a paid license for commercial use (Section 2.1); smaller users may use it free of charge.

Access: You must accept the license on Hugging Face before downloading.

Compiled by this site’s code from the published formats and the license field. License summaries are not legal advice — check the publisher’s original terms before relying on them.

Related Articles

What to Read Next

Sources

Image, Video and AudioComfyUI,IC-LoRA,Lightricks,Lightricks/LTX-2.3,Lightricks/LTX-2.5-22b-IC-LoRA-Restore,LTX-2.3,safetensors,Verified,動画生成