It's a web app that uses the device camera and can capture a direct image or an inverted image, which auto-downloads to the device. Can be used for digitising negatives with a smartphone, although probably not great quality.

Try it out at https://luigiman86.github.io/negativecamera/

I mean, I'm not a dev, I have no idea what I am doing, but I started with the the basic code from this gist: https://gist.github.com/rand0mC0d3r/442ab777ad0bf2c91b6e67f6681a27d0, thanks to @rand0mC0d3r, and rounded it out into a web app, because my friend needs to digitalise negatives. Can you help me do things like make it auto crop? AI colour correction? Let the pull requests flow!

How it works

Everything happens on your device. There is no server, no account, and nothing is uploaded.

• 📹 Open WebCam — starts the camera.
• 🌓 Toggle negative — inverts the preview.
• 📷 Capture — saves a PNG of the current frame, inverted if the toggle is on.
• 🔄 Switch camera — front/back on a phone, cycles cameras on a desktop.
• 🔦 Torch — appears only if your camera has one. Even light matters more than resolution when scanning a negative.
• 🌙 Dark/light — follows your system setting until you override it.

Hover a button for its tooltip, or tap it on a phone for the same hint. The small badge in the top-left shows the camera resolution on the left, the size of the saved image in the middle, and which camera is currently in use.

Notes on the implementation

A few things that are less obvious than they look:

• The saved file matches the preview exactly. The preview is a centre-crop of the camera frame (to fill the screen), and the capture applies that same crop at the camera's native resolution. You frame the shot, and you get that shot.
• Inversion is applied to a static snapshot, not with a canvas filter. The canvas filter property is not available in iOS Safari, so using it silently saved un-inverted negatives on iPhones. The capture now copies the frame into a snapshot canvas with a plain draw, then inverts it with a difference blend against white — that is exactly invert(100%), works everywhere, and is pixel-exact. Doing it this way avoids a multi-megabyte getImageData round trip, which is slow on a 4K frame.
• Each capture uses a fresh canvas. Reusing one canvas means the PNG encoder can still be reading it while the next capture resizes it. There is also a toDataURL fallback in case a browser never calls the toBlob callback, so a capture is never silently lost.
• Captures go through toBlob, not toDataURL. A 4K frame as a base64 data URL is tens of megabytes of string, which can crash the tab on a phone.
• Filenames are negative-YYYYMMDD-HHMMSS-NNN.png — no colons, which are illegal in filenames on Windows, and a counter so two captures in the same second do not collide.
• The service worker precaches the app shell, so it opens offline once installed. It is registered as ./service-worker.js so it works in a subdirectory as well as at a domain root.
• Tooltips are drawn by the app, not the browser. A single shared tooltip can only hold one piece of text, so it ends up showing whichever control was used last; and the native title attribute has the same problem because the browser keeps one tooltip slot per page and will redisplay a stale label. Each control now gets its own tooltip, positioned above it and kept inside the viewport.
• The camera dropdown is allowed to be wide. It is the only control that shows text, so it takes up to 460px rather than being squeezed to a fraction of the bar.

Still to do

Ideas, in rough order of usefulness:

• Auto crop — detect the film edge automatically and crop to it.
• Colour correction — a per-channel curve to remove the orange base of the negative and neutralise the cyan cast that inversion leaves behind. A black/white point picker plus three sliders gets most of the way there without any machine learning.
• Drag-to-crop — a manual crop rectangle, as a fallback while auto-crop is being tuned.
• Aperture priority — a simple brightness/contrast preview so you can check the exposure before capturing.

Local development

The camera needs a secure context, so file:// will not work. Serve the folder over HTTP:

python -m http.server 8000

then open http://localhost:8000. On a phone, use your computer's LAN address and make sure the browser is on https or localhost — iOS will not grant camera access over plain http on a LAN address.