The Fast GAN Video Upscaler workflow is designed to efficiently upscale videos using a Generative Adversarial Network (GAN) model, specifically Real-ESRGAN. This workflow is ideal for enhancing video resolution, making it suitable for applications where higher quality is desired without the need for creating new details, as diffusion models might. The workflow utilizes several key nodes, including the UpscaleModelLoader and ImageUpscaleWithModel, to apply the upscaling model to each frame of the video. It begins by loading your video with the LoadVideo node, breaking it down into frames with GetVideoComponents, and then applying the upscaling process.

Technically, the workflow leverages the Real-ESRGAN model to upscale each frame by a factor determined by the model (e.g., 2x or 4x). This is achieved through the ImageUpscaleWithModel node, which processes each frame individually before the frames are reassembled into a video using the CreateVideo node. Finally, the SaveVideo node outputs the upscaled video. This process is notably faster than diffusion-based methods, making it a practical solution for quick enhancements. For optimal results, it is recommended to first upscale the video to the highest desired resolution and then resize it to the target resolution if needed.