Video Upscale
A simple upscaling app for your videos. Uses the open source model SeedVR2 to sharpen your clips.
Try an example
About this workflow
This ComfyUI tutorial walks you through a lean, reliable pipeline to upscale and sharpen existing videos using the open-source SeedVR2 model. The workflow loads your source clip with VHS_LoadVideo, reads technical details like frame rate and resolution via VHS_VideoInfo, and breaks the clip into frames with GetVideoComponents. You can quickly iterate on a small section using Video Slice, then set your target output size with ResizeImageMaskNode.
Each frame is enhanced by the SeedVR2 inference node (3531d186-daf3-492a-bb11-94a4c791a3f3) to recover detail and improve perceived sharpness. Finally, CreateVideo reassembles the processed frames at the original FPS from VHS_VideoInfo, and SaveVideo writes the result to disk. The result is a pragmatic, easy-to-tune upscaler you can use for everything from legacy phone footage to screen recordings without extra complexity or proprietary dependencies.
Questions & answers
VHS_LoadVideo relies on ffmpeg, so common formats like MP4, MOV, MKV, and WEBM typically work. SaveVideo outputs a standard encoded video file. If you encounter format issues, transcode your source to H.264 MP4 and try again.
This minimal pipeline focuses on video frames; CreateVideo/SaveVideo outputs are usually silent by default. If you need audio, extract and remux it externally or extend the workflow with audio-capable nodes before final export.
You can set any width/height in ResizeImageMaskNode, but quality and VRAM usage scale with resolution. A 2× upscale is a safe starting point; pushing beyond 4× can increase artifacts and processing time.
Use Video Slice to test short segments, lower the target output resolution in ResizeImageMaskNode, and process long videos in multiple passes. Close background apps to free GPU memory if you’re near your VRAM limit.