How to Upscale a Photo (Make a Low-Resolution Image Bigger)

A small photo is too small for what you need it for. Stretching it just makes it blurry. AI-powered upscaling can intelligently add detail, producing larger images that actually look sharper. Here’s how it works and what to use.

Why basic resizing doesn’t help

Traditional image resizing (the kind that happens when you stretch an image in Word or any image tool) does bilinear or bicubic interpolation — mathematically calculating in-between pixels.

The result: the image is technically larger but no actual detail is added. Smooth areas stay smooth; edges become slightly blurry; you don’t suddenly see more detail.

Math can’t create information that wasn’t there.

How AI upscaling works

AI upscaling is different. It uses neural networks trained on millions of image pairs (low-resolution + high-resolution versions of the same scene). The network learns to predict what high-resolution detail “should” look like for typical image patterns.

When you feed it a low-resolution photo, it generates plausible high-resolution detail — not by inventing arbitrary information, but by recognizing patterns from its training and applying them.

Result: a larger image that looks sharper, even though some of the “detail” is AI-generated rather than original.

When AI upscaling works well

Photos of people: faces, hair, clothing — AI has trained on millions of these and produces good results.

Nature: trees, grass, water, skies — recognizable patterns AI handles well.

Architecture: buildings, edges, geometric features.

Anime/manga and illustrated art: specific AI models (Real-ESRGAN with anime model, waifu2x) specialize in this and produce excellent results.

When AI upscaling doesn’t work as well

Text and signage: AI can “invent” plausible-looking but wrong characters. Real text from a photo becomes gibberish.

Highly specific patterns: a unique tattoo, an unusual textile, anything outside common patterns.

Heavily compressed sources: starting from a tiny, heavily-compressed JPG limits what AI can recover.

Extremely aggressive upscaling (8× or 16×): even AI hits limits.

Free AI upscaling tools

Upscale.media (free with watermark, paid for clean):

Upload photo
Choose 2× or 4× upscaling
Download result

Bigjpg (free with limits):

Up to 16× upscaling
Free tier has wait times

ImgUpscaler (free with limits): similar service.

Topaz Photo AI / Gigapixel (paid, $99-200): industry standard, excellent quality.

Real-ESRGAN (free, open source, requires Python):

pip install realesrgan
realesrgan-ncnn-vulkan -i input.jpg -o output.jpg -s 4

For local/private upscaling without uploading: Real-ESRGAN is the right answer.

Phone apps

Remini (free with paid tier): popular for photo restoration. Good for old/blurry portraits.

Photo Enhancer (various apps): mobile-friendly AI upscaling.

Adobe Photoshop Express (free): has Super Resolution feature with paid Adobe account.

For occasional phone-based upscaling: Remini works well.

Workflow: how to upscale

Identify what you want: 2×, 4×, or larger upscaling
Pick a tool: free with watermark, paid for clean, or local install
Upload (or load locally)
Choose upscaling level — 2× is usually safer and faster; 4× is more dramatic
Download/save result
Compare to original — verify quality is acceptable

What gets exaggerated

AI upscaling tends to exaggerate certain things:

Sharpness: results often look “too sharp” — over-corrected. Look unnatural compared to original.

Skin: faces can look unnaturally smooth or “plastic.”

Edges: contrast at edges often boosted.

Color: sometimes over-saturated.

For best results: use lower upscaling factors (2× rather than 4×) when possible. Use professional tools that have multiple model options.

Comparison: typical results

Starting from a 400 × 300 portrait photo:

Basic resize to 1600 × 1200: blurry, no added detail.

AI 4× upscale to 1600 × 1200: visibly sharper than basic resize. Eyes, hair, skin detail more visible.

AI 8× upscale to 3200 × 2400: dramatic improvement; high resolution suitable for prints.

Professional Topaz Gigapixel 8× upscale: high quality; sometimes barely distinguishable from original-resolution photo.

For meaningful upscaling: 4× is typical. 8× is aggressive. 16× is for specific cases (old family photos, etc.).

Use cases

Restoring old family photos: 1980s 320 × 200 scanned photos become viewable on modern screens.

Phone-shot photos for print: a 1024 × 768 phone photo from 2008 becomes printable at larger sizes.

Increasing thumbnail quality: 200 × 150 thumbnails become more usable.

E-commerce product photos: stretching budget photos to high-resolution requirements.

Stock photos at higher resolution: when you bought/found a small version of an image you need bigger.

Memes and image macros: bigger versions of internet finds.

What you can’t do

Recover full detail from very low resolution: a 50 × 50 thumbnail can become bigger but won’t suddenly look like a 4K image.

Read text that’s lost to compression: AI guesses; results aren’t accurate text.

Create realistic faces from blur: extreme blur isn’t recoverable.

Match original capture quality: AI inferences differ from real captured detail.

For uses requiring true detail: find a higher-resolution source instead of upscaling.

Cost considerations

One-off upscaling: free tools with watermarks usually work for personal use.

Many images: subscription tools (Upscale.media monthly) or paid software (Topaz Gigapixel one-time purchase) for batch.

Privacy-sensitive content: local tools (Real-ESRGAN) avoid uploading.

After upscaling

Compare critically to ensure quality is acceptable
Crop to focus on the best parts: Image Cropper
Resize to specific target: Image Resizer
Compress for sharing: Image Compressor

TL;DR

Basic resizing: doesn’t add detail
AI upscaling: uses neural networks to plausibly invent detail
Best free tools: Upscale.media, Bigjpg, Real-ESRGAN (local)
Paid standards: Topaz Photo AI, Gigapixel
2× is safe; 4× is good; 8×+ is aggressive
Works well for: faces, nature, architecture, art
Works poorly for: text, unique patterns, heavily compressed sources
After: Image Cropper and Image Compressor for final cleanup