This node provides an easy-to-use interface for the STAR video super-resolution model within ComfyUI. STAR can upscale low-resolution videos by 2x-4x while preserving and enhancing details, with specialized models for light and heavy degradation scenarios.

Features

Two Degradation Models: Choose between Light and Heavy degradation models optimized for different video quality levels
Flexible Upscaling: Use target Resolution for upscaling.
Text Prompting: Guide the enhancement with text descriptions
Advanced Sampling: Multiple samplers (heun, dpmpp_2m_sde) and solver modes
Automatic Model Download: Models are automatically downloaded from Hugging Face Hub on first use
ComfyUI V3 Integration: Uses comfy_api.latest, io.ComfyNode, and ComfyExtension
Cache-Friendly Modular Workflow: Model loading, preparation, text encoding, sampling, VAE decoding, and color correction are independent nodes

Installation

Install xformers before installing this custom nodes please check xformers version which match with currnet pytorch at https://github.com/facebookresearch/xformers/releases .
Navigate to your ComfyUI custom nodes directory:

cd ComfyUI/custom_nodes/

Clone this repository:

git clone https://github.com/vjumpkung/ComfyUI-STARWrapper.git

Install dependencies:

cd ComfyUI-STARWrapper
pip install -r requirements.txt

Restart ComfyUI

Alternative Install

Search in ComfyUI-Manager name ComfyUI-STARWrapper then click install.

Usage

For new workflows, use the nodes under video/upscaling/STAR:

Load Images ──> STAR Prepare Video ───────────────────┐
                                                    v
STAR Model Loader ──> STAR Text Encode ──────> STAR Sample
        │                                           │
        └────────────────────────────────────> STAR VAE Decode
                                                    │
Original Images ─────────────────────────────> STAR Color Fix ──> Images

Configure the nodes as follows:

Connect the source IMAGE batch to STAR Prepare Video and set the target short-side resolution.
Select the degradation checkpoint and precision in STAR Model Loader.
Connect the model to STAR Text Encode and enter the prompt.
Connect the prepared video, model, and conditioning to STAR Sample, then choose the sampling settings.
Connect the sampled latent and model to STAR VAE Decode.
Connect the decoded frames and original source frames to STAR Color Fix.

The released STARVSRNode ID remains available as the deprecated STAR Video Super Resolution (All-in-One) V3 node, so existing workflows continue to load. New workflows should prefer the modular graph.

Why the nodes are split

| Changed value | Nodes that can remain cached | | --- | --- | | Prompt | Model loader and video preparation | | Seed, steps, CFG, sampler, solver mode, or temporal chunk length | Model loader, video preparation, and text encoding | | VAE decode chunk size | Everything through diffusion sampling | | Color-fix method | Everything through VAE decoding |

The prepared video, prompt conditioning, and sampled latent are stored on CPU between nodes to avoid reserving VRAM merely for caching.

VAE encoding remains part of STAR Sample on purpose. The STAR VAE samples its latent distribution and shares the seeded random-number stream with diffusion; splitting that boundary would change seeded results and make execution-order effects easier to introduce. Temporal chunk calculation also stays inside the sampler because it is cheap and has no reusable standalone value.

Main sampling settings

Configure STAR Sample with:

Model: Choose "Light Degradation" or "Heavy Degradation" in the loader
Prompt: Describe the desired output (e.g., "a high quality video") in the text encoder
Resolution: Set the target short side (720p, 1080p, 2160p, ...) in the preparation node
Steps: Number of denoising steps (15-50 recommended)
CFG: Guidance scale (7.5 default)
Sampler: Choose the sampling method
Max Chunk Length: For long videos, process fewer frames per temporal window (32 default)

The final node outputs upscaled video frames as a standard ComfyUI IMAGE batch.

Parameters

| Parameter | Description | Default | Range | | ------------- | ----------------------------- | ----------------- | ----------------- | | model | Degradation type | Light Degradation | Light/Heavy | | prompt | Text guidance for enhancement | "a good video" | - | | resolution | Target Resolution | 720 | 16-16384 | | max_chunk_len | Maximum frames per chunk | 32 | 1-128 | | cfg | Guidance scale | 7.5 | 0.0-20.0 | | sampler | Sampling method | dpmpp_2m_sde | heun/dpmpp_2m_sde | | solver_mode | Solver speed | fast | fast/normal | | steps | Denoising steps | 15 | 1-100 | | seed | Random seed | 42 | 0-2^64 | | vae_decode_chunk | Frames per VAE decode pass | 1 | 1-8 | | color-fix method | Output color correction | AdaIN | AdaIN/wavelet/none |

Requirements

Python 3.10+
PyTorch with CUDA support
xformers (CUDA13 is not working)
A current ComfyUI build that provides the V3 comfy_api.latest backend API
See requirements.txt for full dependencies

Models

Models are automatically downloaded from the SherryX/STAR Hugging Face repository:

Light Degradation: I2VGen-XL-based/light_deg.pt
Heavy Degradation: I2VGen-XL-based/heavy_deg.pt

Downloaded models are cached in ComfyUI's models/STAR/ directory.

Credits

This is a ComfyUI wrapper for the original STAR project:

STAR (Spatial-Temporal Augmentation with Text-to-Video Models for Real-World Video Super-Resolution)

Original Repository: https://github.com/NJU-PCALab/STAR
Developed by: NJU-PCALab (Nanjing University)
Paper: STAR (Spatial-Temporal Augmentation with Text-to-Video Models for Real-World Video Super-Resolution)

Please cite the original work if you use this in research:

@misc{xie2025starspatialtemporalaugmentationtexttovideo,
      title={STAR: Spatial-Temporal Augmentation with Text-to-Video Models for Real-World Video Super-Resolution}, 
      author={Rui Xie and Yinhong Liu and Penghao Zhou and Chen Zhao and Jun Zhou and Kai Zhang and Zhenyu Zhang and Jian Yang and Zhenheng Yang and Ying Tai},
      year={2025},
      eprint={2501.02976},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2501.02976}, 
}

Third-Party Components

This wrapper also uses:

Color correction adapted from sd-webui-stablesr by Li Yi
Stable Video Diffusion VAE from Stability AI
OpenCLIP text encoder

License

This wrapper follows the license of the original STAR project. Please refer to the original repository for licensing details.

Troubleshooting

CUDA Out of Memory

Reduce max_chunk_len to process fewer frames at once
Lower the target resolution.
Process shorter video segments

Model Download Issues

Ensure you have internet connection for first-time model download
Check Hugging Face Hub accessibility
Models are ~2-3GB each, ensure sufficient disk space

Color Artifacts

The node automatically applies AdaIN color correction
Try adjusting the prompt for better color guidance
Experiment with different CFG values

Contributing

Contributions are welcome! Please ensure:

Code follows the existing style (uses ruff for formatting)
Test with both degradation models
Update documentation for new features

Support

For issues related to:

This ComfyUI wrapper: Open an issue in this repository
The STAR model itself: Refer to the original STAR repository