Six real photo tools. Zero uploads.
Upscale, crop for a passport, pull text out of an image, translate it, restore an old print, or clean up a product shot — all processed on your own device.
Tools
What PixKit can't do (on purpose)
Removing a background from any photo, swapping clothes, fixing a face, erasing an object, or generating a cartoon/headshot all need a trained AI model running on a server — that means an account, a cost per image, or both. PixKit only includes tools that genuinely work for free, in your browser, with real math and image processing rather than a fake or watered-down version of an AI feature. See the guide for details on each tool's technique.
🔍 Upscale Image
Enlarge a photo 2× or 4× with sharpening to counter softness.
Upscale Image
High-quality resampling plus an unsharp-mask pass to keep edges crisp after enlarging.
How this upscaler works
Enlarging an image means inventing pixels that were never in the original — there's no getting around that. This tool uses the browser's own high-quality image resampling (the same technology under professional image editors' basic resize) to scale up cleanly, then applies an unsharp mask: a technique that compares the enlarged image to a blurred version of itself and pushes edges further apart to restore some of the crispness that scaling naturally softens.
What this isn't
True AI super-resolution models are trained on millions of image pairs to hallucinate plausible fine detail that was never captured — hair strands, fabric texture, and so on. That requires a server-side neural network. This tool is a real, useful enlargement technique, but it won't invent detail beyond what a good resize plus sharpening can recover.
Other tools
🪪 Passport Photo
Crop and resize to standard passport/visa dimensions.
Passport Photo
Crops around the head area and resizes to standard passport/visa dimensions at print-ready DPI.
Getting the dimensions right
Passport and visa photos are rejected constantly for being the wrong physical size or resolution. This tool crops to the exact aspect ratio your destination requires, then resizes to the precise pixel dimensions for a standard print DPI (300 by default), so the file prints at the correct physical size without any guesswork.
What you still need to check yourself
Requirements vary by country and change over time: background color (usually plain white or off-white), head size as a percentage of the frame, neutral expression, no glasses glare, and photo recency are all things this tool doesn't verify. Always check your destination's current official requirements before submitting.
Other tools
📝 Extract Text (OCR)
Pull the text out of a photo or screenshot.
Extract Text (OCR)
Reads text out of a photo or screenshot using on-device OCR (Tesseract.js) — no server involved.
On-device OCR
This tool uses Tesseract.js, an open-source optical character recognition engine compiled to run directly in your browser via WebAssembly — the same underlying engine (Tesseract) used by many commercial OCR products, running entirely on your device rather than uploading your image to a server.
What affects accuracy
Clear, high-contrast, horizontally-aligned text works best. Handwriting, stylized fonts, low resolution, and skewed or rotated text all reduce accuracy — if results look off, try cropping tighter to just the text region and using a higher-resolution photo.
Other tools
🌐 Translate Image
Extract text from an image, then translate it.
Translate Image
Extracts text from the image on-device, then opens Google Translate with that text pre-filled — translation itself happens on Google's site, not locally, since a real free on-device translator for arbitrary languages isn't practical to bundle.
Extract, then hand off to translate
This tool runs the same on-device OCR as Extract Text, then opens Google Translate in a new tab with the extracted text already filled in. A genuinely free, accurate, offline translator covering dozens of languages isn't something that can be bundled into a browser tab — real machine translation models are large and compute-intensive. Handing off to an existing, free translation service is the honest way to complete this without pretending to run a translation model that isn't actually there.
Why not translate silently in the background?
Any free translation API would need a key and a cost, or would be relying on scraping a service in a way that breaks easily and violates its terms. Opening Google Translate directly, with your text pre-filled, is transparent about where the translation is actually happening.
Other tools
🧡 Restore Old Photo
Auto color, contrast, and de-yellowing for aged prints.
Restore Old Photo
Automatic histogram-based color correction, contrast, mild denoising, and de-yellowing for aged or faded prints.
What "restoration" means here
Old prints fade unevenly, yellow with age, and pick up grain or speckling. This tool runs an automatic levels correction (stretching each color channel's actual range back out to full contrast), a de-yellowing pass that counters the amber cast common in aged photos, a contrast boost, and a light denoise that smooths speckling without erasing real detail.
What this isn't
It won't repair a physical tear, reconstruct a missing corner, or remove a deep scratch through the image content — that requires generative inpainting, a neural network task. This restores color and clarity; it doesn't invent missing content.
Other tools
📦 Product Photo
Auto-crop and clean up the background for listings.
Product Photo
Auto-crops around your product and whitens a plain/uniform background — works best on a solid-color backdrop.
How the background cleanup works
This tool samples the four corners of your photo to estimate the background color, then finds every pixel that differs meaningfully from that color to build a bounding box around your subject. It crops to that box with some padding, then blends any remaining near-background-colored pixels toward pure white — a technique called color-distance keying.
When it works well, and when it doesn't
This works nicely for a product shot on a plain, evenly-lit seamless background (white, gray, or a solid color) where the subject's colors are clearly different from the backdrop. It won't cleanly separate a subject from a busy, textured, or unevenly-lit background — that's a job for real subject segmentation, which needs a trained model.
Other tools
How PixKit works, honestly
What each tool actually does under the hood, and why some AI photo-editing features aren't included.
Why some tools aren't here
- Remove background (any photo), change clothes, fix a face, remove an object, replace a background, turn into a cartoon, and generate a professional headshot all require a trained AI model — segmentation, generative inpainting, or style transfer — running on a server with real compute cost per image. Offering a fake or badly-degraded version of these would be worse than not offering them. PixKit only includes tools that are genuinely free, run in your browser, and actually work.
Technique by tool
| Tool | Technique |
|---|---|
| Upscale | High-quality canvas resampling + unsharp mask sharpening |
| Passport photo | Aspect-ratio crop + resize to exact print-DPI pixel dimensions |
| Extract text | Tesseract.js — on-device OCR via WebAssembly |
| Translate image | Same on-device OCR, then hand-off to Google Translate |
| Restore old photo | Auto levels (histogram stretch) + contrast + de-yellow color balance + light denoise |
| Product photo | Corner-sampled background color + color-distance keying + auto-crop |
Privacy
Every tool processes your image using the HTML canvas, directly in your browser. Nothing is uploaded to a server. The Extract Text and Translate Image tools download a small open-source language model file from a CDN the first time you use them (needed for the OCR engine itself) — your image is still never sent anywhere.