Short version: the picture gets chopped into a grid of little squares, and each square is replaced by the character that looks most like it. Everything else is detail.
Your image gets shrunk down to a grid — say 173 squares across. That number is the Detail slider. More squares means more characters, so finer things survive; fewer squares means a chunkier, more abstract picture.
Characters are taller than they are wide, so the grid is stretched to match. That's why the result doesn't come out squashed.
Each square gets averaged down to a single brightness value. Bright squares and dark squares are what the character choice is based on.
Before that, the brightness is stretched so the darkest part of your image counts as black and the brightest as white. A washed-out photo would otherwise only ever use the middle few characters and come out as grey mush.
Characters hold different amounts of ink. @ fills a lot of space; . barely any. A dark square gets a heavy character, a bright one gets a light character.
Rather than trusting a hard-coded order, every character you pick is actually drawn and measured first. So you can paste in whatever symbols you like and they still come out correctly ordered from light to dark.
Each character takes the colour of the pixels it actually covers, not the average of the whole square. On a face against a dark background, that's the difference between skin tones and mud.
Or throw the colour away entirely — green-on-black, two-tone, or a fixed palette like Game Boy.
- \ | /.Raise Contrast, or switch to a character set with more steps — Detailed has 70 characters where Classic has 10.
Raise Detail and try Shape mode. A little Edge strength brings back eyes and mouths.
Adjust Character aspect under Characters. It tells the converter how tall your font's characters really are.
Turn on dithering, or raise its amount.
Pick a Destination. Chat and Terminal cap the width for you and show the character count.
Raise Flicker reduction. If it's still noisy, add Frame blending.
The text converter doesn't analyse a picture at all. It uses banner fonts — files that store a hand-drawn ASCII picture of every letter — and glues those letters together, overlapping them so they interlock instead of sitting in separate boxes. There are 328 of them here, plus one-line styles that swap your letters for look-alike characters so the result stays a single line.
Nowhere. There is no server and no upload. Your file is read straight off your disk into the browser tab, converted there, and forgotten when you close it.