⚑ SeedVR2 Image Upscaler in Google Colab

This repository contains an easy-to-use Google Colab notebook for running the SeedVR2 Image Upscaler. Powered by ComfyUI, this tool allows you to take any image and upscale it to a stunning high resolution using the SeedVR2 DiT (3B) model, entirely for free.

πŸŽ₯ Watch the Tutorial: Setup SeedVR2 Image Upscaler in Google Colab

πŸš€ Run in Colab: Open Google Colab Notebook

Open In Colab

Get Pro


✨ Features Supported in this Notebook

This notebook automates the upscaling process into 3 simple steps:

  1. βš™οΈ Initialize Core Environment: Automatically clones ComfyUI and installs the custom SeedVR2 Processing Nodes.
  2. πŸ“₯ High-Speed Asset Downloader: Uses Aria2c to rapidly download the SeedVR2 DiT (3B) and VAE models in the background.
  3. πŸ–ΌοΈ SeedVR2 Image Upscaler: The core upscaling engine. Features include:
    • Source Image Upload: A built-in prompt allows you to upload your image directly into the Colab environment.
    • Upscale Factor: A slider to multiply the shortest edge of your image by 1.0x up to 4.0x.
    • Color Correction: Options to apply lab color correction or none.
    • Upscale Alpha: Built-in support for upscaling images with transparency (RGBA formats).
    • Noise Controls: Fine-tune the INPUT_NOISE_SCALE and LATENT_NOISE_SCALE.
    • Auto-download: Automatically downloads the final high-resolution image directly to your computer.

πŸ› οΈ How to Use

  1. Click the "Open in Colab" badge above.
  2. Go to Runtime > Change runtime type and ensure a T4 GPU is selected.
  3. Run Cell 1 to initialize the environment and install dependencies.
  4. Run Cell 2 to download the required DiT and VAE models.
  5. Go to Cell 3.
    • Ensure UPLOAD_INPUT_IMAGE is checked.
    • Adjust your UPSCALE_FACTOR and other settings as desired.
    • Hit the Play button and upload your image when prompted.
    • The server will process your image, display a preview, and automatically download the upscaled version!

🀝 Credits

Downloads last month
-
Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support