Upscale Any Image to 4K for Free with Upscayl

Upscayl is a free, open-source AI image upscaler that runs entirely on your GPU — no subscriptions, no cloud uploads, no watermarks. It uses state-of-the-art Real-ESRGAN and custom-trained neural networks to enlarge photos, artwork, screenshots, and textures while intelligently reconstructing detail.

In 2026, Upscayl has become the default tool for anyone who wants professional-grade AI upscaling without paying for services like Topaz Gigapixel or Adobe Super Resolution. This guide covers installation, model selection, batch processing, and getting the best results for different types of images.

Why AI Upscaling Beats Traditional Methods

Traditional upscaling (bicubic, Lanczos) simply interpolates between existing pixels, creating blurry, blocky results at large magnifications. AI upscaling works differently:

A neural network trained on millions of image pairs learns what high-resolution details should look like
It predicts and synthesizes missing detail rather than guessing from neighbors
The result looks sharper and more realistic, especially at 4x and 8x magnification

The difference is dramatic: a 512×512 image upscaled to 2048×2048 with AI looks nearly as sharp as a native 2K image. Traditional methods produce visible blur and artifacts at the same scale.

Installation

Upscayl is available for Windows, macOS, and Linux as a desktop app.

Windows

Download the .exe installer from upscayl.org or the GitHub releases page. Run the installer and launch Upscayl from the Start Menu.

macOS

Download the .dmg file. Open it, drag Upscayl to Applications. On first launch, right-click and select Open to bypass Gatekeeper.

Linux

AppImage (universal):

wget https://github.com/upscayl/upscayl/releases/latest/download/upscayl-x.x.x-linux.AppImage
chmod +x upscayl-*.AppImage
./upscayl-*.AppImage

Flatpak:

flatpak install flathub org.upscayl.Upscayl
flatpak run org.upscayl.Upscayl

Snap:

sudo snap install upscayl

GPU Requirements

Upscayl uses Vulkan for GPU acceleration, which means it works with:

NVIDIA GPUs (GTX 1000 series and newer)
AMD GPUs (RX 500 series and newer)
Intel Arc GPUs
Apple Silicon (M1/M2/M3/M4 via Metal)

On machines without a supported GPU, it falls back to CPU — much slower but functional.

The Upscayl Interface

The main UI has three panels:

Left panel — settings: scale factor, model selection, output format
Center panel — image preview with before/after comparison slider
Right panel — file browser for batch mode

Workflow:

Select an image or folder
Choose a model
Choose scale (2x, 3x, 4x, or custom)
Set output folder
Click Upscayl

Upscaling Models

Upscayl ships with several built-in models and supports custom model imports.

Built-in Models

Model	Best For	Speed
General Real v3	Photos, everyday images	Fast
General Photo v3	Portrait and landscape photography	Medium
Digital Art	Illustrations, concept art, anime-style	Fast
Realesrgan x4plus	Versatile general purpose	Medium
Realesrgan x4plus Anime	Anime, manga, cartoon art	Fast
Remacri	Real photos with fine detail	Slow
UltraMix Balanced	Photos where sharpness/softness balance matters	Medium
UltrasharpV10	Maximum sharpness, can over-sharpen	Fast

Which Model to Use

Content Type	Recommended Model
Landscape/nature photos	General Photo v3
Portrait photos	UltraMix Balanced or General Photo v3
Anime / manga / illustrations	Realesrgan x4plus Anime or Digital Art
Game screenshots / UI	General Real v3
Old scanned photos	Remacri
Artwork / concept art	Digital Art
Text and diagrams	UltrasharpV10

Adding Custom Models

Upscayl supports any .param + .bin model pair in the ncnn-vulkan format.

Installing Custom Models

In Upscayl, go to Settings → Model Path
Note the models directory (e.g., ~/.config/upscayl/models/ on Linux)
Download a custom model — the community maintains models at upscayl/upscayl-models

# Example: download a community model
cd ~/.config/upscayl/models/

wget https://github.com/upscayl/upscayl-models/raw/main/models/lollypop.param
wget https://github.com/upscayl/upscayl-models/raw/main/models/lollypop.bin

Restart Upscayl — the model appears in the model list automatically

Popular Community Models

Model	Specialty
Lollypop	Painted/artistic photos with soft detail
SwinIR	General purpose, academic benchmark leader
HAT	Highest quality single-image super-resolution
DRCT-L	Fast and high quality

Batch Processing

Process an entire folder at once using Upscayl’s batch mode.

Via the UI

Click Batch Upscayl in the top menu
Select your input folder
Choose model and scale
Select output folder
Click Upscayl

Upscayl processes images sequentially (one at a time on GPU). A folder of 100 images at 4x might take 5–30 minutes depending on GPU speed and image resolution.

Via CLI (upscayl-bin)

Upscayl also ships a CLI tool called upscayl-bin (actually realesrgan-ncnn-vulkan under the hood):

# Find the binary inside the Upscayl installation
# Linux AppImage: extract with --appimage-extract first
# Windows: C:\Program Files\Upscayl\resources\

# Single image upscale
./upscayl-bin \
  -i input.jpg \
  -o output.png \
  -n realesrgan-x4plus \
  -s 4 \
  -f png

# Batch folder upscale
./upscayl-bin \
  -i ./input_folder/ \
  -o ./output_folder/ \
  -n realesrgan-x4plus \
  -s 4

# List available models
./upscayl-bin -h

Output Formats and Quality Settings

Upscayl supports PNG, JPG, and WEBP output.

Format	Notes
PNG	Lossless, largest files — best for further editing
JPG	Lossy, smaller files — set quality to 95+ to minimize artifacts
WEBP	Good compression with quality control

For photography workflows, export as PNG first, then convert to your target format with ImageMagick or GIMP to maintain full quality through any post-processing.

Getting the Best Results

Pre-processing Tips

Denoise before upscaling if the source is noisy. Noise gets upscaled too. Use Topaz DeNoise, Neat Image, or GIMP’s G’MIC denoise filters first.
Crop tightly to your subject before upscaling to reduce processing time.
Avoid double-upscaling — running an image through Upscayl twice rarely helps and can introduce artifacts.

Scale Factor Tips

2x is fastest and often sufficient for printing at 2x the original print size
4x is the sweet spot — best quality-to-processing-time ratio
Custom scale (e.g., 1.5x or 3x) is available for precise output dimensions

When AI Upscaling Struggles

Severely compressed images (heavy JPEG artifacts) — the AI will upscale the artifacts too
Very blurry images — AI adds detail that may not be accurate
Images with text — try UltrasharpV10 model for better text clarity

Integrating Upscayl into Workflows

Automatic1111 / ComfyUI + Upscayl

For Stable Diffusion workflows, generate at 512×512 or 768×768 for speed, then run Upscayl afterward to reach 2K or 4K output quality. This is faster than generating at native high resolution.

GIMP Script-Fu Batch

# Batch upscale using upscayl-bin, then load results in GIMP
for f in *.jpg; do
  ./upscayl-bin -i "$f" -o "upscaled_${f%.jpg}.png" -n realesrgan-x4plus -s 4
done

Upscayl vs. Alternatives

Tool	Price	Offline	Quality	Speed
Upscayl	Free	Yes	Excellent	Fast (GPU)
Topaz Gigapixel	$199	Yes	Best-in-class	Fast
Adobe Super Res	CC Sub	Yes	Very Good	Medium
Let’s Enhance	Subscription	No	Good	Cloud
Waifu2x	Free	Yes	Good (anime)	Medium

Upscayl sits comfortably between free tools like Waifu2x and paid tools like Topaz — it’s genuinely excellent and completely free.

Final Thoughts

Upscayl is one of those rare tools that delivers professional results for zero cost. Whether you’re restoring old family photos, upscaling game screenshots, preparing artwork for print, or enhancing surveillance footage, Upscayl handles it all locally and privately.

Download it, spend five minutes choosing the right model for your content type, and you’ll have a tool you reach for constantly.