floyo logo
Powered by
ThinkDiffusion
floyo logo
Powered by
ThinkDiffusion

ComfyUI-SeedVR2_VideoUpscaler

2098

Last updated
2025-12-24

ComfyUI-SeedVR2 Video Upscaler is an advanced tool designed for enhancing video and image quality through high-quality upscaling. It integrates seamlessly with ComfyUI, leveraging state-of-the-art diffusion models to ensure temporal consistency and support for various formats.

  • Supports both RGB and RGBA formats, allowing for sophisticated handling of videos with transparency.
  • Features advanced memory optimization techniques like BlockSwap and VAE tiling, enabling operation on systems with limited VRAM.
  • Offers a modular architecture with dedicated nodes for different processing stages, enhancing workflow control and efficiency.

Context

This tool is an official implementation of the SeedVR2 model tailored for ComfyUI, aimed at providing users with high-quality video and image upscaling capabilities. It utilizes advanced machine learning techniques to improve the visual fidelity of media, making it suitable for various applications in creative and professional environments.

Key Features & Benefits

The SeedVR2 Video Upscaler boasts several practical features, including:

  • High-Quality Upscaling: Using diffusion-based techniques, it enhances image and video quality significantly, preserving details and reducing artifacts.
  • Temporal Consistency: This feature ensures that the upscaled frames maintain coherence across the video, which is crucial for smooth playback.
  • Flexible Resolution Handling: It supports upscaling to any resolution divisible by two, accommodating various output needs without compromising quality.

Advanced Functionalities

The tool includes advanced functionalities such as:

  • BlockSwap Technology: This allows for dynamic swapping of transformer blocks between GPU and CPU memory, enabling the processing of large models on systems with limited VRAM.
  • VAE Tiling: This feature processes large images in smaller tiles, significantly reducing memory usage during encoding and decoding phases.
  • Multiple Model Support: Users can choose from various model configurations (3B and 7B parameter models) depending on their hardware capabilities and quality requirements.

Practical Benefits

By integrating this tool into their workflows, users can expect improved efficiency and quality in their media processing tasks. The advanced memory management techniques help prevent out-of-memory errors, allowing for smoother processing of high-resolution videos. Additionally, the modular architecture enhances user control over the upscaling process, making it easier to customize settings for specific needs.

Credits/Acknowledgments

This project is a collaborative effort by NumZ and Adrien Toupet from AInVFX, based on the original SeedVR2 model developed by the ByteDance Seed Team. Acknowledgments are also due to various community contributors who have provided improvements and bug fixes, enhancing the tool's functionality and reliability. The code is released under the Apache 2.0 license, promoting open-source collaboration and usage.

Inner Nodes

SeedVR2LoadDiTModel, SeedVR2LoadVAEModel, SeedVR2TorchCompileSettings, SeedVR2VideoUpscaler