Enlarge images and videos with classic or local AI upscaling

Platform note

What Upscaler is for

Upscaler is useful when you want to enlarge a photo, graphic, screenshot, scan, video, or other supported media file. It can use normal mathematical scaling methods or, where supported, local AI models that attempt to reconstruct detail.

• Enlarge supported image files.
• Enlarge supported video files where available.
• Choose scale amounts for resizing.
• Use classic scalers such as Nearest, Bilinear, Bicubic, and Lanczos.
• Use RealESRGAN local AI upscaling where supported.
• Choose model types such as photo, general, or anime/line-art style models where available.
• Use tiling/performance options for large files where available.
• Batch process files after testing settings on one file.
• Save enlarged copies without overwriting originals.

Classic upscaling vs. local AI upscaling

• Classic upscaling: Resizes pixels mathematically. It is faster, predictable, and good for simple enlargement.
• Local AI upscaling: Uses a local model such as RealESRGAN where supported. It can add perceived detail and reduce some artifacts, but results vary by image.

How to enlarge an image or video with classic upscaling

1. Open Upscaler.
   In Grafiq, open More → Upscaler.
2. Add files.
   Click + Add Files… and choose the image or video files you want to enlarge.
3. Choose an output folder.
   Click Choose Output Folder…, or use the default output location if you are testing.
4. Set the scale.
   Choose how much larger you want the output to be. For example, 2× doubles width and height.
5. Choose a classic method.
   Select a method such as Nearest, Bilinear, Bicubic, or Lanczos.
6. Test one file first.
   Process one representative file before running a large batch.
7. Start upscaling.
   Run the upscale and check the exported result.

How to use RealESRGAN/local AI upscaling where supported

1. Open Upscaler.
   Open More → Upscaler.
2. Open AI Upscaler.
   Choose AI Upscaler (RealESRGAN)… if it is available in your build.
3. Add files.
   Click + Add Files… and choose the images or videos you want to process.
4. Choose AI scale.
   Pick an available scale such as 2× or 4×, depending on the model.
5. Choose a model.
   Use a photo model for natural photos, a general model for mixed content, or an anime/line-art model for illustrated content where available.
6. Use tiling/performance options if needed.
   For very large files, use conservative tiling options for reliability.
7. Preview if available.
   Preview still images when supported. Video preview may not be available in some workflows.
8. Start RealESRGAN.
   Run the upscale and check the exported result.

Which upscaling method should you choose?

• Nearest: Best for pixel art or hard-edged graphics when you want crisp, blocky enlargement.
• Bilinear: Fast and softer. Useful for quick previews or simple resizing.
• Bicubic: Balanced and generally useful for normal photos.
• Lanczos: Sharper classic resizing. Good for high-quality enlargement, but may create halos on some edges.
• RealESRGAN Photo: Best for natural photos where supported.
• RealESRGAN General: Good for mixed content where supported.
• RealESRGAN Anime/Line Art: Best for illustrated content, line art, and animation-style images where supported.

Good uses for Upscaler

• Enlarge old or small photos for slideshows or documents.
• Make small images easier to reuse in projects.
• Improve the perceived sharpness of supported images with local AI where available.
• Scale up screenshots, scans, graphics, or image assets.
• Upscale video files where supported.
• Try classic scaling when you want predictable resizing.
• Try local AI upscaling when you want stronger reconstruction and are willing to experiment.

Tips for better upscaling results

• Start from the best original file you have.
• Upscale once from the original instead of repeatedly upscaling the same exported file.
• Test one file before batch processing.
• Use 2× before trying 4× if the result looks artificial or too processed.
• For huge files, use conservative tiling/performance options where available.
• Use classic Lanczos for predictable photo enlargement.
• Use Nearest for pixel art or retro graphics.

What to watch out for

• Upscaling does not magically restore perfect original detail.
• AI upscaling results vary by image, model, scale, and source quality.
• Some AI results may look too smooth, too sharp, or artificial.
• 4× upscaling creates much larger files and can take longer.
• Video upscaling can take significantly longer than image upscaling.
• RealESRGAN/local AI upscaling is not available on Intel Macs in the Mac App Store build.
• Keep originals until you confirm the exported result is useful.