THE LATENT IMAGE

Film emulation has a credibility problem. Most tools apply a warm curve, desaturate the shadows slightly, and call it Portra. It looks fine. It doesn't look like film.

I've been shooting film long enough to know the difference, and building software long enough to want to do something about it.

Latent started as a question: what would it take to emulate a film stock properly? Not aesthetically — technically. What does Kodak Portra 400 actually do to light?

The answer turns out to be three things. A tone curve for each channel that captures how the film responds to exposure — the shadow lift, the midtone character, the way highlights compress before they clip. A 3×3 colour matrix that models cross-channel contamination — Fujifilm's green bleed into skin, Kodak's warm red channel that makes everything feel like late afternoon. And per-hue corrections that target specific colour zones — the way Portra flattens and warms skin, the way Velvia pushes greens and blues independently of everything else.

That model gets baked into a 33×33×33 3D LUT and runs in WebGL in your browser. No server. No upload. Your photo never leaves your device.

The privacy angle wasn't the starting point — it was a consequence of taking the technical approach seriously. If the processing happens locally, there's nothing to send. So we don't send it.

What it isn't

The honest part of building this is knowing where the models are wrong.

The gold standard for film emulation is shooting a colour reference card on the actual stock, having it lab-developed and professionally scanned, and fitting your model to the measured data. That process costs time and money. I haven't done it for everything yet.

Stocks marked ✦ Certified in Latent have been validated against a real scan. Everything else is a well-researched model built from published sensitometry, manufacturer data, and years of looking very closely at film photographs. They're good. They're not measured. I'm honest about that on the site because you deserve to know.

Where it's going

The thing I keep coming back to is how much knowledge lives inside film photographers that doesn't exist anywhere useful. If you've shot Portra obsessively for a decade, you know things about how it renders that aren't in any datasheet. That knowledge stays in your head, or in a forum thread from 2014.

Latent is an attempt to build a place for that knowledge to land. A contribution model where if you have reference data — a .cube LUT built from a real scan, a colour card shot on a stock I haven't certified yet — you can submit it, I'll review it, and your name goes on the work.

latent.photos — 22 stocks, free, no account.