AI Image Upscaler

Enlarge photos 2x–4x with a real neural network — in your browser, free, unlimited, private.

100% private — your photos never leave your device

Upscale your image

Drop an image here or click to browse Best for photos and screenshots up to ~2 MP. AI runs on your device.

Add an image to begin
A compact Real-ESRGAN model (~5 MB) loads once, then upscales tile by tile.

How it works

1

Add your image

Photos, screenshots, old scans — anything up to about 2 megapixels. It never leaves your device.

2

AI rebuilds the detail

A Real-ESRGAN neural network processes your image in tiles, reconstructing edges and textures that simple resizing just blurs. Takes seconds to a couple of minutes depending on your device.

3

Download the result

Crisp PNG at 2x or 4x the original size. Unlimited images, no watermark, no account.

Guides for This Tool

How to Upscale an Image With AI: 2x vs 4x, Print Sizes and Honest Limits

2x vs 4x explained, the 1200×900 to 4800×3600 print example (16×12 inches at 300 DPI), and the honest limits — blur, tiny text and waxy textures included.

Upscaling only works on files your other tools can read. If the photo is still in Apple's format, run it through the HEIC Converter first — JPG or PNG out, then upscale. The step-by-step version, with the batch settings and quality choices, is our guide to converting HEIC to JPG in bulk.

An upscaled portrait often ends up on a document. Before you print, check it against the passport photo guide — official sizes, head-height rules and the print layout that passes. And for finished video work, the Video QC Checker scans for black frames, frozen frames and dead audio before the upload goes public.

Frequently asked questions

Honestly: no. Topaz uses much larger models and costs $149/year for a reason. This uses a compact 5 MB Real-ESRGAN model that runs in a browser tab — the results genuinely beat every free web upscaler and crush plain resizing, which is remarkable for free. For professional print work, Topaz remains the tool.
The AI always runs at 4x internally (that's what the model does); the 2x option simply downsamples that result, which is faster overall because fewer output pixels are written. If you're in a hurry, 2x still looks dramatically better than the original.
Yes — you're running a neural network on a phone CPU. Recent Chrome/Edge on desktop with WebGPU is several times faster. Keep the tab in the foreground until it finishes. Warmth during 4x jobs is normal and harmless; if the phone gets hot, pause between images and let it cool rather than forcing a batch through.
No. The model file lives on this site and all inference happens in your browser via ONNX Runtime. Nothing leaves your device. That includes the original photo and the enlarged result: both are processed and downloaded locally, so private pictures stay private even on a shared computer once you close the tab.
Yes — the general x4v3 model handles illustrations well, sharpening lines and flat colors nicely. Dedicated anime models exist but this one is a strong all-rounder. It also copes with pixel art and screenshots, though very small faces in a busy scene are where any upscaler, this one included, starts to guess.
JPEG, PNG, and WebP all work — anything your browser can decode. The output is always PNG, because re-saving as JPEG would throw away some of the detail the AI just created. Transparency in PNG sources is preserved through the upscale.
Only partly. If the photo is slightly soft, the AI reconstruction can make it look noticeably crisper. But true motion blur or out-of-focus shots stay blurry — the model sharpens edges, it cannot recover information the lens never captured. For a blurry original, expect “better” rather than “fixed.”

Worked example: from phone photo to an 8×10 print

Say you have a 1200×900 phone photo of your dog — fine on a screen, but only 4×3 inches at a decent 300 DPI print. You want an 8×10 framed print. Drop the photo into the upscaler, choose 4x, and the tool rebuilds it at 4800×3600 pixels. At 300 DPI that covers a 16×12 inch print, so your 8×10 crop has real detail instead of stretched pixels.

Processing takes longer at 4x because the model is inventing sixteen times the pixels, not four — each output pixel is analyzed in context. On a mid-range laptop expect under a minute for 2x and a few minutes for 4x. The result downloads as a PNG you can hand to any print shop, and because everything ran locally, the original never left your device.

Limitations

Upscaling is educated guessing, not magic. The AI reconstructs plausible detail, so tiny text in the background may come back looking sharp but wrong — always zoom in and check faces, signs, and text before printing. It cannot fix motion blur or heavy JPEG artifacts; those get amplified into confident-looking smudges.

Very large source images combined with 4x can exhaust your browser’s memory and crash the tab — if that happens, try 2x first, or downscale the source slightly. Results also vary by subject: portraits and illustrations upscale beautifully, while dense foliage or fabric textures sometimes turn waxy. Treat the preview as the verdict, not the promise.

Related Tools

Passport Photo Maker — Free ID Photos & Print

Make compliant passport & ID photos at home: country size presets, head-size guides, 4×6 print sheet. Free — no $9…

Beat Tag Dropper

Drop your producer tag over any beat automatically: set the interval, fades and ducking, preview the protected beat…

Dialogue Analyzer

Paste a story chapter for a free dialogue report: dialogue-vs-narration ratio, speech-tag census, adverb check and…

Media Metadata Scrubber

Strip EXIF, GPS, IPTC and XMP metadata from photos and videos in one click. Batch scrubbing with before/after reports…

Lower Third Maker

Design broadcast-style lower thirds free — 5 presets, custom colors and fonts, safe-margin guides, transparent PNG…

Video Inventory — Metadata Spreadsheet

Audit up to 100 video files in your browser: duration, resolution and size in a sortable table with CSV export…