Video2X: Machine learning video upscaling and interpolation
Video2X enhances video quality using machine learning algorithms for upscaling, frame interpolation, and restoration with multiple backend support.
Learn more about video2x
Video2X is a comprehensive video enhancement toolkit that leverages state-of-the-art machine learning models to upscale, interpolate, and restore video content. It supports multiple processing backends including Real-ESRGAN, Anime4K, RIFE, and libplacebo for various enhancement tasks. The framework provides both CLI and GUI interfaces, enabling batch processing with hardware acceleration through NVIDIA CUDA, AMD ROCm, and Vulkan. Video2X handles complex video processing pipelines including codec management, audio preservation, and subtitle handling. It's particularly effective for anime content upscaling, real-world video enhancement, and frame rate interpolation up to 60fps or higher. The tool supports containerized deployment and includes comprehensive preprocessing options for optimal results across different video types.
Multiple AI Backend Support
Video2X integrates multiple cutting-edge machine learning models including Real-ESRGAN for photorealistic upscaling, Anime4K for anime-optimized enhancement, and RIFE for frame interpolation. Users can select the most appropriate algorithm for their specific content type, ensuring optimal quality enhancement whether working with anime, live-action footage, or mixed media content with flexible configuration options.
Hardware-Accelerated Processing
Leverages GPU acceleration across NVIDIA CUDA, AMD ROCm, and Vulkan backends to dramatically reduce processing time for large video files. The framework automatically optimizes memory usage and batch processing for maximum throughput, making it feasible to upscale hours of content efficiently. Multi-threading support and intelligent queue management ensure optimal hardware utilization throughout the enhancement pipeline.
Production-Ready Video Pipeline
Provides complete video processing infrastructure with automatic codec handling, audio stream preservation, subtitle management, and metadata retention. The system includes robust error handling, resume capability for interrupted processing, and comprehensive logging. Container support enables deployment in automated workflows, while the CLI interface facilitates integration into existing media processing pipelines with scriptable batch operations.
from video2x import Video2X
upscaler = Video2X()
upscaler.process(
input_path="input.mp4",
output_path="output_upscaled.mp4",
scale_factor=2
)Linux binaries now ship as AppImage; older RIFE models (pre-v4.6) excluded from downloads and must be fetched manually.
- –Download older RIFE models manually from the repo's models/rife directory if you need versions before v4.6.
- –Videos without PTS information are now supported; metadata copying from input to output streams is enabled.
Patch release fixing PTS precision degradation in long videos and Qt6 config restoration bugs.
- –Upgrade to resolve PTS timestamp drift that worsens with video duration, preventing sync issues in long encodes.
- –Note that only RIFE v4.6 and v4.0 ship in downloads; manually fetch older models from the repo if required.
Adds Real-CUGAN ncnn Vulkan support and a unified logger; Qt6 GUI gains in-app logs, French translation, and update checks.
- –Enable Real-CUGAN ncnn Vulkan processing for GPU-accelerated upscaling alongside existing RIFE models.
- –Note encoder now always uses calculated PTS with corrected math, fixing potential timestamp drift issues.
Related Repositories
Discover similar tools and frameworks used by developers
bolt.new
LLM-powered browser IDE with integrated WebContainers runtime.
text-generation-webui
Feature-rich Gradio-based UI for running and interacting with LLMs locally, supporting multiple model formats and extensions.
ultralytics
PyTorch library for YOLO-based real-time computer vision.
Mask2Former
Unified transformer architecture for multi-task image segmentation.
nanoGPT
Minimal PyTorch implementation for training GPT models.