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: Support for 2x, 3x, and 4x upscaling factors
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 Integration: Seamless integration with ComfyUI workflows

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

In ComfyUI, find the node under video/upscaling → STAR Video Super Resolution
Connect your video frames (as IMAGE tensors) to the input
Configure the parameters:
- Model: Choose "Light Degradation" or "Heavy Degradation"
- Prompt: Describe the desired output (e.g., "a high quality video")
- Resolution: Target Resolution (720p, 1080p, 2160p, ...)
- Steps: Number of denoising steps (15-50 recommended)
- CFG: Guidance scale (7.5 default)
- Sampler: Choose sampling method
- Max Chunk Length: For long videos, process in chunks (32 default)
The node outputs upscaled video frames

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

Requirements

Python 3.8+
PyTorch with CUDA support
xformers (CUDA13 is not working)
ComfyUI
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 ./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 upscale factor
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