Neura Market
Upscale-A-Video logo

Upscale-A-Video

Paid

Temporal-consistent AI video upscaling with diffusion models.

5.0
Image EditingContact
Inputs: video, textOutputs: video
Type
Saas
Visit Website

About Upscale-A-Video

Upscale-A-Video is a diffusion-based model for real-world video super-resolution, presented at CVPR 2024 as a Highlight. Developed by researchers at Nanyang Technological University, it enhances video resolution while maintaining temporal consistency across frames. The model accepts low-resolution videos and optional text prompts to guide the upscaling process. It includes support for color correction (AdaIn, Wavelet), optional integration with LLaVA for automatic prompt generation, and provides a command-line inference script with tunable parameters such as noise level, guidance scale, and diffusion steps. The repository also supplies pretrained models and the YouHQ dataset for training and evaluation.

Key Features

Diffusion-based video super-resolution with temporal consistency
Accepts low-resolution video and optional text prompts as inputs
Color fix options: AdaIn and Wavelet
Optional LLaVA for automatic prompt generation
Tunable parameters: noise steps, guidance scale, inference steps
Supports AIGC videos, old movies, animations, and more
Provides pretrained models and the YouHQ dataset (38,576 training videos, 40 test clips)

Pros & Cons

Pros
  • Open-source and free to use
  • Published as a CVPR 2024 Highlight, indicating strong peer-reviewed research
  • Specifically designed for temporal consistency, reducing flickering across frames
  • Supports text prompts for guided upscaling
  • Flexible color correction options to match output colors
Cons
  • Requires manual setup: cloning repo, creating conda environment, installing dependencies
  • Needs downloading large pretrained models before first use
  • Optimized for GPU usage; may not run efficiently on CPU
  • Command-line interface only; no graphical user interface or web service
  • Inference time can be long for high-resolution or many frames

Best For

Upscaling AI-generated content (AIGC videos)Restoring and enhancing old movies, animations, and vintage footageImproving video quality for archival or production purposes

Alternatives to Upscale-A-Video

Final2x

Open-source AI image lossless upscaling tool with multiple models

Petalica Paint

Transform sketches into vibrant art effortlessly with Petalica Paint.

Variart

Transform copyrighted images into unique, copyright-free versions with Variart.

Recolorai

Unleash Your Creativity with ReColor AI

Secta Labs

Secta Labs advanced AI image tool converts selfies into professional headshots within hours with a selection of styles available in a private gallery.

Upscayl

Upscale low-res images 8x with stunning colors, details, and contrast, all for free.

FAQ

What are the system requirements for Upscale-A-Video?
The tool requires Python 3.9, a conda environment, and the dependencies listed in requirements.txt. It is designed for GPU acceleration (recommended).
How do I run inference on a video?
Use the inference script with the input video path. Example: python inference_upscale_a_video.py -i ./inputs/video.mp4 -o ./results -n 150 -g 6 -s 30. Additional parameters like prompt and color fix are available.
Does Upscale-A-Video support text prompts?
Yes, the model accepts text prompts as inputs to guide the upscaling process. Prompts can be provided via the command line.
Can I use it for old movies or animations?
Yes, the repository includes example commands specifically for old videos, movies, and animations.