ComfyUI-LTXVideo: LTX-2 Nodes, Workflows and IC-LoRAs for ComfyUI
LTX-Video Support for ComfyUI
At a glance
- What is it?
- Lightricks' ComfyUI-LTXVideo adds LTX-2 nodes and example workflows on top of the model already built into ComfyUI core. It is aimed at CUDA users with 32GB+ VRAM, and its IC-LoRAs cover control, HDR, dubbing and upscaling rather than plain text-to-video.
- Who is it for?
- Adopt ComfyUI-LTXVideo if you already run ComfyUI on a CUDA GPU with at least 32GB of VRAM and 100GB of free disk, and you want LTX-2 control, HDR, dubbing or upscaling graphs rather than plain text-to-video. Skip it if your GPU is below that memory bar, if you want a standalone CLI or Python API, or if a pixel-accurate upscaler is what you need, because the Pixel Spatial Upscaler re-renders detail instead of interpolating.
- Can I use it commercially?
- Check first. The repository uses a licence we do not classify automatically, so read its LICENSE file before any commercial use.
- Is it still maintained?
- Yes. The repository last received commits 3 days ago.
- What is it written in?
- Mainly Python, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on November 15, 2026, and from our analysis. They are not legal advice.
Star history
- Lightricks/ComfyUI-LTXVideo4,268 ★ · +130
| Repository | 2026-09-23 | 2026-11-15 |
|---|---|---|
| Lightricks/ComfyUI-LTXVideo | 4.1k | 4.3k |
130 Stars in the last 53 days, measured from the daily GitHub snapshot. Compare with other projects
Editorial analysis
What ComfyUI-LTXVideo adds on top of ComfyUI core
LTX-2 is already built into ComfyUI core, at comfy/ldm/lightricks, so a stock ComfyUI install can run the model. This repository is the extra layer: custom nodes plus example workflows for the features core does not ship. The README describes it as "a collection of powerful custom nodes that extend ComfyUI's capabilities for the LTX-2 video generation model." The audience is narrow and stated plainly in the prerequisites: a CUDA-compatible GPU with 32GB+ VRAM and 100GB+ free disk space for models and cache. That is a workstation or rented-cloud profile, not a laptop one. The value is in what the nodes cover beyond base generation. The repository adds IC-LoRA variants for control signals, HDR output, speech dubbing and spatial upscaling, plus samplers and loaders that appear as separate files in the repository root (tiled_sampler.py, low_vram_loaders.py, looping_sampler.py, q8_nodes.py). If you only want text-to-video from LTX-2, core ComfyUI is enough and this extension is optional.
How the nodes and LoRAs are wired together
The repository is a flat set of ComfyUI node modules registered through nodes_registry.py, with __init__.py as the entry point ComfyUI loads. Functionally the code splits into a few groups. Conditioning and text handling live in conditioning_loader.py, gemma_encoder.py and gemma_api_conditioning.py, so prompts can be encoded locally or through a Gemma-based API path with system_prompts/ and gemma_configs/ supplying the configuration. Sampling and decoding are separate concerns: easy_samplers.py, tiled_sampler.py, looping_sampler.py and decoder_noise.py on one side, tiled_vae_decode.py on the other. The IC-LoRA work sits in iclora.py and iclora_attention.py. HDR is its own pipeline: hdr_nodes.py, hdr_color.py and hdr_io.py. The README describes the Union IC-LoRA as a single model that handles depth and edge control conditions, and states that it "operates on a downsampled latent size, which reduces memory usage and significantly speeds up inference while maintaining quality." Note the trade-off implied there: control is applied at reduced resolution, so fine edge alignment is not the goal. The HDR path is similarly explicit about data flow. The LoRA produces frames in LogC3-compressed space, and the LTXVHDRDecodePostprocess node decodes them back to linear HDR values, emitting both a Reinhard-tonemapped SDR preview and the raw linear HDR tensor.
Installing ComfyUI-LTXVideo through Comfy Manager
The README's Quick Start recommends Comfy Manager rather than a manual git clone. Open ComfyUI, click the Manager button or press Ctrl+M, choose Install Custom Nodes, search for "LTXVideo", click Install, wait, then restart ComfyUI. The nodes then appear in the node menu under the "LTXVideo" category, and the README states that required models are downloaded on first use. If you prefer to inspect the package first, requirements.txt lists what gets pulled in:
colour-science>=0.4.4
diffusers
einops
huggingface_hub>=0.25.2
kornia
ninja~=1.11.1.4
openimageio
transformers>=4.50.0Two entries there are not optional conveniences. The comments in requirements.txt say colour-science is required for HDR primaries with "no baked-matrix fallback", and openimageio is required for HDR EXR read and write with "no OpenCV EXR fallback". If you only plan to run the non-HDR workflows you can still install the full set, but be aware that the HDR feature will not silently degrade if those packages are missing. After restarting, the first real step is to load a workflow instead of building a graph by hand. The example workflows ship inside the installation directory:
ComfyUI/custom_nodes/ComfyUI-LTXVideo/example_workflows/For LTX-2.3 the README lists single-stage and two-stage distilled graphs, an IC-LoRA union control graph for depth, human pose and edges, a motion-tracking graph, HDR, Dub-It and pixel spatial upscaling, plus a text-to-audio graph. Pick the file whose name matches your intent, load it in ComfyUI, and the LTXVideo nodes should already be resolved. The older 2.0 workflows remain in example_workflows/2.0/ for text-to-video, image-to-video, a video-to-video detailer and the earlier IC-LoRA graphs.
Where ComfyUI-LTXVideo runs into trouble
The prerequisites are the first wall. A CUDA-compatible GPU with 32GB+ VRAM rules out most consumer cards, and 100GB+ of free disk is a real constraint once model weights and cache accumulate. The repository does contain low_vram_loaders.py and a tiled sampler, which suggests the authors are aware of memory pressure, but the prerequisites still state the 32GB+ figure. The second limitation is architectural. This is a ComfyUI extension, not a library. The README's Quick Start drives everything through the node graph and the Manager install flow. If your pipeline is headless and scripted, you are adopting ComfyUI as well as this repository. Third, the IC-LoRAs are not general-purpose refiners. The README is direct about the Pixel Spatial Upscaler: it "synthesizes fine detail rather than simply interpolating pixels" and is "a creative upsampler, not a pixel-accurate refiner." For archival, measurement or restoration work that needs fidelity, that is the wrong tool. Fourth, the HDR path has hard dependencies and a specific colour space. It targets ACEScct working space with LogC3-compressed LoRA output, and the README offers a tiled 2.5 graph or a 2.3 single-stage graph for it. If your pipeline is already in a different colour management scheme, you are converting at the boundary.
ComfyUI-LTXVideo compared with running LTX-2 in ComfyUI core
The honest alternative is not another repository; it is ComfyUI core. LTX-2 lives in comfy/ldm/lightricks upstream, so text-to-video and image-to-video generation are available without installing anything from Lightricks. The difference is scope. Core gives you the model. ComfyUI-LTXVideo gives you the graphs and the special-purpose LoRAs: the Union IC-LoRA that merges depth and edge control into one adapter, the HDR IC-LoRA that outputs linear HDR alongside an SDR preview, the Dub-It IC-LoRA for multilingual dubbing and same-language rephrasing with speaker identity preservation, and the Pixel Spatial Upscaler in 2x and 4x variants. It also adds the plumbing those need, such as iclora_attention.py, hdr_io.py and the tiled decode path. So the comparison is not quality against quality, it is whether your task is covered by base generation. If you are generating a clip from a prompt, core is sufficient and you avoid the 32GB+ VRAM framing that the README attaches to this extension. If you need control signals, HDR deliverables or speech replacement, core does not provide them and this repository is the path the vendor documents.
Maintenance, licence and upgrade cost
The repository is not archived, and the last push was on 2026-10-20, which is recent. The latest GitHub release is v1.1.0, published 2026-10-20, matching the version in pyproject.toml. That matters for upgrade cost. The README distinguishes LTX-2.3 workflows from older LTX-2.0 workflows and offers both a tiled 2.5 HDR graph and a 2.3 single-stage HDR graph, so the example_workflows directory is versioned by model generation. When LTX-2 moves again, expect to migrate graph files rather than bump a pin. Models are downloaded on first use, which means a fresh workflow can pull large weights without an explicit install step, and the 100GB+ disk figure is the buffer for that. On licensing, GitHub reports the licence as NOASSERTION, meaning it could not match the LICENSE file to a known identifier. The LICENSE file exists at the repository root, so read it directly rather than assuming terms from the LTX-2 model repository or from ComfyUI itself. Model weights, the LoRAs and the node code may not all carry the same terms. This is not legal advice; it is a pointer to the one file that can answer the question.
Editorial conclusion
Adopt ComfyUI-LTXVideo if you already run ComfyUI on a CUDA GPU with at least 32GB of VRAM and 100GB of free disk, and you want LTX-2 control, HDR, dubbing or upscaling graphs rather than plain text-to-video. Skip it if your GPU is below that memory bar, if you want a standalone CLI or Python API, or if a pixel-accurate upscaler is what you need, because the Pixel Spatial Upscaler re-renders detail instead of interpolating. Before committing, open ComfyUI/custom_nodes/ComfyUI-LTXVideo/example_workflows/ and confirm the graph matching your GPU and your LTX-2 version is present, and check the LICENSE file in the repository, since GitHub reports the licence as NOASSERTION.
Frequently asked questions
What is LTX in ComfyUI?
LTX-2 is Lightricks' video generation model, and it is built into ComfyUI core under comfy/ldm/lightricks. ComfyUI-LTXVideo is the separate repository that adds custom nodes and example workflows for LTX-2 features beyond base generation.
Can I use ComfyUI to generate video?
Yes. With LTX-2 in ComfyUI core and the LTXVideo nodes installed, the README lists text-to-video and image-to-video workflows under example_workflows, including single-stage and two-stage distilled graphs for LTX-2.3.
How do I install ComfyUI-LTXVideo?
The README recommends Comfy Manager: open ComfyUI, press Ctrl+M, choose Install Custom Nodes, search for "LTXVideo", click Install, then restart ComfyUI. The nodes appear under the "LTXVideo" category, and required models download on first use.
What hardware does ComfyUI-LTXVideo need?
The prerequisites list a CUDA-compatible GPU with 32GB+ VRAM and 100GB+ free disk space for models and cache. The repository also includes low_vram_loaders.py and a tiled sampler.
Does ComfyUI-LTXVideo upscale video accurately?
No, and the README says so. The Pixel Spatial Upscaler "synthesizes fine detail rather than simply interpolating pixels" and is described as a creative upsampler, not a pixel-accurate refiner, with 2x and 4x variants.
What licence does ComfyUI-LTXVideo use?
GitHub reports the licence as NOASSERTION, which means the LICENSE file at the repository root could not be matched to a known identifier. Read that file directly instead of assuming terms from ComfyUI or the LTX-2 model repository.
Official sources
Add this badge to your README
If you maintain this project, the badge below links readers to this analysis and shows its maintenance status from the daily GitHub snapshot. Paste the markdown into your README; add ?metric=license or ?metric=stars to the image URL for a different field.
[](https://hysenlabs.com/projects/lightricks-comfyui-ltxvideo)Star history chart for your README
[](https://hysenlabs.com/projects/lightricks-comfyui-ltxvideo)Was this useful?