How to upscale video to 4K for free with AI

Updated November 2026 · 7 min read

Normal scaling only makes pixels bigger. AI upscaling rebuilds detail the source never stored, so edges, text and textures look sharp at 4K. Here is how to do it for free with FrameScale AI, without uploading your video anywhere.

What AI upscaling actually does

When a player or editor enlarges 1080p to 4K, it interpolates between existing pixels (bicubic or Lanczos). The result is bigger but soft. An AI super-resolution model such as Real-ESRGAN was trained on millions of image pairs, so it predicts plausible high-resolution detail: crisp edges, readable text, clean textures and less blocky compression.

The same frame upscaled by FrameScale AI: sharper edges, texture and detail Low-resolution video frame enlarged with normal scaling BeforeAfter · FrameScale
640-pixel-wide frame enlarged 2× normally (left) vs FrameScale AI with Real-ESRGAN General v3 ×2 at 20% noise removal (right). Footage: Tears of Steel © Blender Foundation, CC BY 3.0.

Which scale to use

SourceScaleResult
1920×1080 (Full HD)2×3840×2160 (4K UHD)
1280×720 (HD)2×2560×1440 (QHD)
960×5404×3840×2160 (4K UHD)
720×576 / 720×480 (DVD, SD)4×about 2880 pixels wide, beyond Full HD

Step by step

  1. Install FrameScale AI. Download the installer for Windows, macOS or Linux from the download section and install it. It is free, needs no account, and runs entirely on your computer.
  2. Add your video. Open FrameScale AI and drag your clip onto Drop videos here. It reads anything FFmpeg can decode: MP4, MKV, MOV, AVI, MPEG-2, ProRes and more.
  3. Turn on Upscale and pick a model. In the pipeline on the right, switch on Upscale. For camera footage choose Real-ESRGAN x4+ (live action) for the best detail, or Real-ESRGAN General v3 (live action, fast) for speed, with a Noise removal slider. For anime and cartoons choose Real-ESRGAN x4+ Anime or Real-ESR AnimeVideo v3 (fast).
  4. Choose 2× or 4×. Use 2× for 1080p to 4K and 4× for low-resolution sources (see the table above).
  5. Preview 2 seconds. Drag the marker under the viewer to a busy scene and click Render 2 s. A progress bar shows the time left; then compare the result split or side by side, and zoom in to check fine detail.
  6. Pick the output format. Under output, choose HEVC 10-bit (best compatibility for 4K) or AV1 10-bit (smallest files) in MP4 or MKV, and a quality level. HDR, audio, subtitles and chapters are carried over automatically.
  7. Render. Click Render. The job is written in 4-second segments, so you can pause, quit or even survive a crash and continue where it stopped.

Tips for the best 4K result

Old tapes? Interlaced VHS and DV need deinterlacing first. See how to restore VHS tapes with AI.

Frequently asked questions

Can I upscale video to 4K for free?

Yes. FrameScale AI is free and open source, with no watermark, no resolution limit and no account. It uses your own GPU, so nothing is uploaded.

Does AI upscaling really add detail?

It reconstructs plausible detail learned from training data, which looks far sharper than normal scaling. It cannot recover information that was never captured, so very blurry sources improve less.

How long does upscaling take?

It depends on the model, resolution and GPU. On a current mid-to-high-end GPU, the fast General v3 model upscales 540p to 4K at roughly 13 frames per second; a 10-minute 1080p video to 4K takes from minutes to a few hours.

What GPU do I need?

Any Vulkan-capable NVIDIA, AMD or Intel GPU, or a Mac with Apple Silicon. 4 GB of video memory is comfortable; it also runs on the CPU, much more slowly.

Try it on your own footage

FrameScale AI is free and open source for Windows, macOS and Linux. No account, no watermark, no upload.