A History of ASCII Art, ANSI, and the Machines That Draw With Text

Jul 8, 26

ASCII art is the practice of making pictures out of nothing but text: letters, digits, and punctuation arranged so that, from a distance, they resolve into an image. It is one of the oldest forms of computer art and one of the most stubbornly alive. It has outlasted the hardware it was born on, survived the arrival of bitmap graphics, migrated across every network humans have built, and is now being poked at by a new kind of author entirely: the language model.

This is a long tour. We will start before computers existed, work through the line-printer and terminal eras, spend real time inside the ANSI artscene and its neighboring text-art traditions, and end on a question that would have sounded absurd twenty years ago: what does it mean that an AI like Claude can be handed a canvas and asked to draw?

Before the computer

The instinct to draw with type is far older than the machines we usually credit for it. Around 300 BCE, Greek poets wrote technopaignia - “pattern poems” whose lines were shaped into wings, eggs, and altars. In 1633 George Herbert printed “Easter Wings” in the physical shape of wings in his collection The Temple. In 1865 the same year Lewis Carroll printed the mouse’s tale in Alice’s Adventures in Wonderland as a curling tail of text, an anonymous engraver rendered a portrait of Lincoln entirely out of the words of the Emancipation Proclamation. In 1918 Guillaume Apollinaire arranged whole poems into pictures in Calligrammes. The idea that arranged characters could be a picture was thoroughly established long before anyone owned a keyboard.

The direct ancestor of ASCII art is typewriter art. Christopher Latham Sholes patented the first practical typewriter and the QWERTY layout in 1867, and within a generation people were making pictures with it. In 1893 a British stenographer named Flora Stacey composed a butterfly on a typewriter - the earliest known example of the form. By 1939 Julius Nelson had published ARTYPING, a proper how-to guide, and through the 1940s and 50s typewriter art flourished, with practitioners like Paul Smith and Alan Riddell. In parallel, radio-teletype (RTTY) operators working in five-bit Baudot code learned to send images down the wire as patterns of overstruck characters - the same trick, over radio.

None of these people had “pixels.” What they had was a fixed grid of monospaced positions and a small alphabet of marks. That constraint - a grid and a limited set of glyphs - is the whole game, and it has not changed since.

The computer era: Bell Labs and the line printer

The move from typewriter to computer happened, as so many things did, at Bell Labs. Working on early IBM mainframes in the first half of the 1960s, Kenneth Knowlton wrote programs that turned photographs into grids of tiny symbols. In 1964 he used an IBM 7094 to convert a photo into a character image, and in 1966 he and Leon Harmon produced “Studies in Perception I,” a reclining nude rendered entirely out of small electronic symbols. It was shown at the Museum of Modern Art in 1968 and is routinely cited as the moment character-based imagery became recognized as art rather than a curiosity of the machine room.

The standard that gave the form its name arrived in 1963, when ANSI published ASCII - the American Standard Code for Information Interchange - a 7-bit character set with 128 code points. That is the raw material: 95 printable characters, from space to tilde.

What turned ASCII into a folk art was the line printer, especially the IBM 1403. Computer centers printed banners, calendars, and famous images - Snoopy, Spock, a chunky Mona Lisa - on green-bar fanfold paper, and the source code got copied from site to site like a chain letter. Shading was achieved through overstriking: printing several characters on the same position so that a region built up more ink and appeared darker. That is the earliest form of the “density ramp” idea we still use today, where characters are ranked by how much of their cell they fill.

Terminals and home computers

When video terminals such as the DEC VT100 appeared in 1978, two things arrived that would define the next decade. The first was ANSI escape codes (later standardized as ANSI X3.64): short sequences beginning with an escape character that let a program move the cursor, clear the screen, and set colors. The second was the realization that if you can move the cursor and set color, you can draw

Home computers shipped their own extended character sets tuned for graphics. The Commodore’s PETSCII and the Atari’s ATASCII each added dedicated line-drawing and block characters, so a kid with a Commodore 64 could compose surprisingly detailed screens without a single bitmap. These sets have their own devoted revival scenes to this day.

ANSI art: color, blocks, and a real scene

Plain ASCII art is monochrome text. ANSI art is its louder, more elaborate cousin, and for roughly fifteen years it was the dominant form of computer folk art in the West.

Three ingredients made ANSI art possible on the IBM PC. First, the ANSI escape codes (interpreted by the ANSI.SYS driver under DOS) gave sixteen foreground and eight background colors, plus cursor control. Second, code page 437 - the original IBM PC character set - included a set of block and shading characters that were essentially low-resolution pixels:

 full: █     dark: ▓     medium: ▒     light: ░     half: ▀ ▄ ▌ ▐

With those four shades and sixteen colors, an artist could build smooth gradients, rounded forms, and shaded portraits out of a text grid. Third, and crucially, there was a distribution network: the Bulletin Board System.

The artscene

From the early 1980s to the mid 1990s, ANSI and ASCII art exploded on BBSes. Sysops commissioned login screens, menus, and “one-liners.” In the pirate and warez subculture, art was effectively currency: the .NFO file that shipped with a cracked program carried the releasing group’s logo, and the quality of that logo signaled the group’s prestige.

Crews formed, competed, and mentored newcomers. The lineage runs from Aces of ANSI Art (founded around 1989-1990) to the two houses that would define the golden age: ACiD Productions - ANSI Creators in Demand, founded in 1990 - and iCE (Insane Creators Enterprise), founded in 1991. From roughly 1994 to 1999 these groups and others (Dark Illustrated among them) released monthly artpacks: ZIP archives of new .ANS files, viewed in dedicated players. It was a genuine culture, with recognizable personal styles, house aesthetics, rivalries, and stars.

The tooling

ANSI art had real tools. TheDraw was the classic DOS editor. ACiDDraw and later PabloDraw (a cross-platform editor that even supported collaborative, networked drawing) gave artists proper canvases, color palettes, and character browsers. In 1994 a metadata standard called SAUCE (Standard Architecture for Universal Comment Extensions) was introduced so a file could record its title, author, group, and canvas dimensions - the artscene equivalent of EXIF.

Decline and preservation

The graphical web and the GUI eventually pulled attention away from text-mode art. BBSes faded; the artpacks slowed. But the scene was unusually history-conscious, and its work has been preserved with care. Jason Scott’s textfiles.com archived enormous quantities of BBS-era text; 16colo.rs became the canonical archive of ANSI artpacks; and modern groups such as Blocktronics have kept new high-craft ANSI art coming into the 2010s and beyond. Unlike most digital folk art, the ANSI scene largely saved itself.

Neighboring traditions

ASCII and ANSI are the best-known text arts, but they are part of a wider family.

(^_^)     (T_T)     \(^o^)/     (._.)     ¯\_(ツ)_/¯

Emoji descend from this instinct; the shrug survives in group chats worldwide.

The web and chat era

As the internet opened up, text art moved onto Usenet - the group alt.ascii-art - and into email signatures, where a few clever lines were a personal flourish that would survive any mail client. The most influential figure of this era was Joan G. Stark, who signed her work jgs. She was both a prolific artist and the form’s unofficial historian; she called ASCII art “non-graphical graphics” and ran a gallery on GeoCities from 1996. A good number of the pieces in this gallery carry her signature, and a 2023 tribute typeface, the Jgs Font, revived her drawing style for the web.

The era also produced enduring tools. FIGlet (Glenn Chappell and Ian Chai, 1986) turned ordinary words into big banner headings and is still used in terminals today. cowsay (Tony Monroe, 1999) wrapped your text in a speech bubble beside an ASCII cow and became a Unix in-joke that never died. And AAlib (Jan Hubicka, 1997) converted images and video into text in real time - most famously powering Simon Jansen’s Star Wars Asciimation, an attempt to redraw the entire film in characters, frame by frame.

The modern era

ASCII art never really left; it moved into the places where text still lives. Roguelike games like NetHack and Dwarf Fortress render entire worlds in glyphs, where an ampersand is a demon and a colon is a lizard. Terminal tools like neofetch greet you with your operating system’s logo in ASCII. Editors such as JavE and, more recently, ASCII Draw Studio keep the craft properly tooled. Image-to-ASCII converters - jp2a, chafa, libcaca - are a single command away, and chafa in particular uses Unicode and truecolor to produce results that would have stunned a 1990s BBS artist.

Memes carried the form into everyday chat: copypasta built from characters, and the shrug, ¯\_(ツ)_/¯, typed thousands of times a day by people who have no idea it is descended from Japanese kaomoji. And a distinctly contemporary aesthetic has emerged around liminal and “backrooms” text art - empty rooms, endless corridors, and humming fluorescent grids suggested in a handful of characters, where the sparseness is the point.

ASCII art and AI: the Claude question

Here is where the story turns strange. For sixty years, the two ways to make good character art were to draw it by hand or to convert an image algorithmically. The conversion route is a direct descendant of Knowlton’s Bell Labs work: sample the brightness of each cell and map it onto a ramp of characters ordered from empty to solid.

 .:-=+*#%@

That ramp is deterministic and, frankly, unbeatable for photographic subjects. It is exactly the technique this project’s image-import feature uses in the browser.

Then large language models arrived, trained on oceans of text that happened to include a great deal of ASCII art - from alt.ascii-art, from source-code comments, from README files. So a model like Claude has, in a sense, “read” the entire folk tradition. Ask it for a cat and it will often produce a creditable one. But there is an important and honest caveat: language models are uneven at ASCII art, and for a specific, structural reason.

A model generates text one token at a time, left to right, and it never sees the two-dimensional result while producing it. ASCII art is fundamentally spatial - a character on line 5 has to line up with a character on line 9 - but the model is working in a one-dimensional stream, with no rendered canvas to check against. Tokenization makes this harder: runs of spaces and box characters get split into tokens in ways that have little to do with visual columns. The result is that models are excellent at recalling pieces they effectively memorized, and much weaker at composing novel art with correct alignment. The prettiest new ASCII still comes from converters, from human artists, or from the deliberate minimalism of the liminal style, precisely because those routes do not depend on a model holding a picture in its head one token at a time.

The interesting frontier, then, is not “can the AI draw a pretty cat.” It is whether you can give the model the one thing it lacks: a feedback loop and a real canvas. And that is exactly what modern agent tooling makes possible.

Anthropic introduced the Model Context Protocol (MCP), an open standard that lets an AI client call external tools. This project ships an MCP server that exposes a small ASCII drawing engine - commands like draw_line, draw_rect, draw_text, render, and save_to_gallery. With it, Claude does not have to imagine the whole picture in advance. It can start a canvas, issue drawing commands to an actual grid, call render to look at what it has made, correct itself, and only then publish. The agent decides the sequence; the server just executes and reports back. In other words, we hand the model the missing dimension - a place to draw and a way to see the result - and let it work the way a human artist does: iteratively, with the drawing in front of them.

That reframing is the honest and, I think, more hopeful version of “AI ASCII art.” The value is not in a model that has memorized a butterfly. It is in a collaborator that can pick up tools, use a feedback loop, and add to a shared wall - sitting in the same long line that runs from Flora Stacey’s typewriter, through Knowlton’s line printer and the ANSI artpacks, to a gallery that anyone, human or agent, can draw into.

Why it endures

ASCII art survives because of its constraints, not despite them. A piece is just text. It copies and pastes cleanly, renders in any monospace font, needs no image format, and works on hardware separated by half a century. It asks the maker to suggest a great deal with very little, which is a good description of art in general. That is also the idea behind this project: a place to browse the form, draw your own in the editor, convert an image, read its history, and - now - let an AI agent take a turn at the wall.


Dates and figures in this piece draw on asciiart.eu’s history of ASCII art, supplemented with additional background on the ANSI artscene, text-art traditions, and the relationship between language models and character art.