Pixelify Products

Pixelify is one site with three surfaces. The two tools share the same rules: your images never leave your device, output is never watermarked, and there is no account to create.

• AI Image Upscaler: 2× / 4× / 8× super-resolution with three neural models and per-image enhancement controls.
• Background Remover: automatic subject segmentation with a full canvas editor on top.
• Blog & Guides: deep-dives on the models, the privacy story and the practical limits of local AI.

Drop in a PNG, JPEG or WebP up to 15 MB and choose how far to push it. Upscaling is tiled, so large images are processed in chunks that fit your GPU memory, and each pass is cancellable mid-run. The result is post-processed through optional enhancement passes and exported losslessly.

Times are typical for a mid-range desktop GPU; the small model is so light it runs comfortably even on WASM-only devices.

• Scales: 2×, 4× and 8× (8× is a two-pass pipeline).
• Presets per model (Standard, Fast, Crisp, Illustration, Photography, Landscape, Portrait, Artwork), each mapping to concrete denoise / sharpen / colour / HDR values.
• Per-image controls: face restoration, noise removal, sharpening, colour enhancement, HDR boost, brightness and contrast.
• Input: PNG, JPEG, WebP up to 15 MB. Output: PNG, WebP or JPEG, no watermark, up to 16,384 × 16,384 px.
• Runs batch-style: queue several images, each processing independently in the background, with progress, cancel and completion chime per image.

The remover uses the Chopper neural segmentation model, an architecture trained specifically to hold on to hair, fur, glass and other semi-transparent edges that colour-key tools destroy. It runs in a background Web Worker, so the page stays responsive while the model loads and segments.

• One-click automatic cutout; no masks or green screen.
• Canvas editor on top: erase/restore brushes, crop, zoom lens, and a full undo/redo history.
• Replace backgrounds with transparent, solid colour, gradient presets, or your own photo, plus drop shadows, outlines and brightness/contrast/saturation adjustments.
• Input: PNG, JPEG, WebP up to 15 MB (drag, paste or picker). Output: transparent PNG, WebP with alpha, or JPG on a solid background.
• Hair- and fur-level precision is the default, not a premium tier.

The blog is where we write at length: how the models work, what super-resolution actually reconstructs versus invents, how WebGPU differs from WASM on this hardware, and how a site with no backend handles privacy. Articles are published in English, French, Chinese, Japanese, Hindi and Russian, with read-time estimates and tag-based search.

• Any modern browser. WebGPU (Chrome/Edge 113+, Safari 17+ over HTTPS) gives full speed; everything else falls back to WebAssembly automatically.
• Firefox works via WASM. It is functional, but expect slower large-image runs.
• Mobile browsers are supported; WebAssembly is used there deliberately, because mobile WebGPU memory limits can crash the tab on large models.
• A desktop-class GPU and 8 GB+ RAM is recommended for 8× runs on large images; output is capped at 16,384 px per side regardless of hardware.
• First visit downloads the model you pick (1.9MB to 67MB, or about 40MB for the remover). Your browser caches it, so subsequent visits start instantly.

If the site ever changes that model, the Terms of Service will be updated before any paid feature takes effect, and this page will say so plainly.