ワークフローに戻る

Fast GAN Video Upscaler

Upload a video to upscale it using a fast GAN model. The output resolution is determined by your chosen model (e.g., 2x or 4x). For best results, upscale first, then resize to your target resolution.

ワークフローをダウンロードCloudで試す

Input

A short low-resolution clip; every frame is upscaled 4x (480p becomes about 1920 px wide).

Run

Output
00:00
Example

例を試す

このワークフローについて

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.

よくある質問