image-upscaler
An image upscaler enlarges a small image to a larger pixel size while attempting to preserve or recover detail — useful when an old photo, a thumbnail, or a downscaled asset is the only source you have but the destination needs a higher resolution (a print, a large web banner, an upscaled video, a presentation slide). The ZTools Image Upscaler offers classical algorithms (bicubic, Lanczos) plus an AI-style detail-aware mode that runs a small neural network in the browser, allowing 2x and 4x upscale, edge-preserving smoothing, and noise / artifact suppression on the way up — all without uploading the original photo to any server.
Use cases
Restoring old family photos
A 600 px scan from a printed photo album. Upscale 4x for printing or framing without the muddy interpolation that classical bicubic produces.
Recovering low-res screenshots
A screenshot taken on an old monitor needs to land on a 4K display. Upscale 2x with edge-aware mode to keep text crisp.
Stock photo enlargement
A licensed image at 1500 px wide needs to fill a 3000 px banner. Upscale 2x; check for artifacts; touch up if needed.
E-commerce product photo standardisation
Some product shots arrive at 800 px; the marketplace wants 2000 px. Batch-upscale the small ones to meet the catalogue standard.
How it works
- Upload image — JPG, PNG, WebP. Recommended source resolution at least 200 px on the short side; below that, results soften noticeably.
- Pick scale — 2x (most common, fastest) or 4x (slower, biggest jump). Custom percentage also supported.
- Choose algorithm — Bicubic (fast, smoother), Lanczos (sharper, slight ringing on edges), AI-style (best detail, slowest).
- Optional cleanup — Pre-clean noise / JPG artifacts before upscale to avoid magnifying them. Optional post-sharpen for crisp output.
- Export — PNG (lossless, larger) or JPG (smaller, lossy). Original aspect ratio preserved.
Examples
Input: 500x500 thumbnail → 2000x2000 (4x, AI-style)
Output: Detail-aware 2000x2000 with cleaner edges than bicubic
Input: 1200x800 photo → 2400x1600 (2x, Lanczos)
Output: Sharp 2x upscale, slight ringing on high-contrast edges
Input: 800x600 screenshot of UI → 1600x1200 (2x)
Output: Upscaled UI with edge-preserved text
Frequently asked questions
Will upscaling restore real detail?
No — it cannot invent information that was never captured. AI-style upscalers infer plausible detail based on training data; they look better but are guesses, not truth. Treat upscaling as visual smoothing, not forensic restoration.
Which algorithm should I pick?
For photographs, AI-style usually wins. For UI screenshots and line art, Lanczos with light sharpen often beats AI. Bicubic is the safe default — fast and predictable.
Why does the upscaled image look muddy?
Source was already noisy or heavily JPG-compressed. Pre-clean before upscaling, or accept that low-quality input cannot be magic-fixed.
What is the maximum size?
Limited by your device memory. A modern laptop comfortably upscales 4 MP images to 4x; phones may stop at 2x for large inputs.
Does it remove watermarks?
No — and please respect copyright. The tool magnifies what is there; if a watermark is present, the upscale shows the same watermark.
Is the model uploaded to a server?
No — the AI mode uses a small ONNX model that runs in your browser via WebAssembly. Source image never leaves your device.
Tips
- Always start by cleaning JPG artifacts and noise — upscaling magnifies them otherwise.
- For prints, upscale to 300 DPI of the printed size, not just "bigger".
- Compare 2x AI vs 2x Lanczos with a quick A/B preview — different photos prefer different algorithms.
- Keep the original. You may want to re-upscale later with a better model.
- After upscaling, run a small unsharp-mask pass for crisper output before export.
Try it now
The full image-upscaler runs in your browser at https://ztools.zaions.com/image-upscaler — no signup, no upload, no data leaves your device.
Last updated: 2026-05-05 · Author: Ahsan Mahmood · Edit this page on GitHub