Self-hosted Real-ESRGAN x4 video upscaler with temporal smoothing (github.com)

A new self-hosted video upscaling tool, powered by Real-ESRGAN x4, has been announced, offering a free and GPU-optimized alternative to the expensive Topaz Video Upscaler. This tool enhances video quality by upscaling each frame four times and applying a temporal smoothing algorithm to reduce flickering effects typically seen in frame-by-frame neural upscalers. By utilizing a 3-tap weighted average among consecutive frames, the tool effectively maintains clarity during fast motion and scene cuts, preserving both detail and smooth transitions without the intensive computation of commercial alternatives. This development is significant for the AI/ML community as it democratizes high-quality video processing, making powerful upscaling technology accessible at no cost. The community can deploy this tool in various applications, from film restoration to game graphics enhancement. Users can run the tool with a CUDA-compatible GPU, with installation instructions clearly outlined, making it easy for developers to integrate it into their existing workflows. The Real-ESRGAN model supports both photorealistic and anime-style upscaling, underscoring its versatility and relevance in diverse creative fields.