Charis Tsevis

Charis Tsevis

@tsevis.bsky.social

Computational artist and designer. Algorithmic mosaics: images built from thousands of pieces. I build my own tools (Mozaix). tsevis.com

93 followers87 following351 posts

Pinned

African Bricks: portraits built from thousands of fragments, inspired by the brick-and-colour patchwork of townships like Soweto, where colour is a declaration, not decoration. From across the room, a face. Up close, a wall of colour. My most collected series. #mosaic #portrait #ArtistsOnBluesky

Work in progress: teaching a machine to read old pages word by word. Every word it finds becomes a tessera for a new mosaic. The Leningrad Codex (1008), Dante's Divine Comedy (Florence, 1481), the first printed Homer (Florence, 1488) and a Latin manuscript of the Epistles. #mosaic #manuscripts

The best part of this GenAI era is that you can build the tools you need yourself. This weekend I needed an ebook converter for my Mac. So I built one: Biblioteca. Native, free, open-source, with OCR for scanned books, Greek included. github.com/tsevis/biblioteca #macos #opensource

African Bricks: portraits built from thousands of fragments, inspired by the brick-and-colour patchwork of townships like Soweto, where colour is a declaration, not decoration. From across the room, a face. Up close, a wall of colour. My most collected series. #mosaic #portrait #ArtistsOnBluesky

Chris Pappas – The Greek Portrait. A century ago his great-grandfather left Greece for New Hampshire. Today Chris runs for the U.S. Senate. I built his portrait from the architecture of Greece: island houses, domes, arches, staircases, columns. From Greece: Καλή επιτυχία, Chris! 🇬🇷🇺🇸

Prosopon: This app uses InsightFace and Apple Vision to align eyes and mouths in various portraits. I have developed it for a personal project. If anyone needs it for any reason, just drop me a line, and I will email the DMG file. #MacApps

PDF is still one of the worst formats for LLMs. PDFs describe pages for humans, not information for machines. So columns get mixed. Tables collapse. Reading order becomes a guess. #Markdown is much better.

To convert my PDFs, I use Marker, the excellent open-source converter by Datalab. I use it so much that I started modifying it. Desktop interface. Real progress. Better batch conversion. Clear failure reports. No broken half-files. Some fixes are going back upstream too. #opensource

You fill a page with 40 sketches, or lay a whole icon set on one artboard. Then comes the tedious part: crop each one, center it, export it, type a filename. Forty times over. Shapearator does that for you. One sheet in, a clean asset library out. #computervision #opensource

I love Adobe Bridge. It's my favorite file browser. But it can't sort images by color. So I built an extension that does. BridgeColorShorter reads each image's actual palette (not just a single dominant pixel), groups your files by color family, and ramps each group from dark to light. #AdobeBridge

Your camera records the light falling on a face. What if it recorded the heat coming out of it? I made an app for that. It is called Heron. Give it a photo. It gives back an image that looks like it was captured by a machine beyond human sight. Thermal camera. X-ray. Night vision. Kirlian plate.

View full thread

Same deal here. Real physics, but the images are art, not real thermography. I am in an experimental mood. I call it a system mosaic. Not one picture, a system that makes pictures, built from models, algorithms, and frameworks instead of glass tiles. Eight instruments. Sixteen palettes.

Your camera records the light falling on a face. What if it recorded the heat coming out of it? I made an app for that. It is called Heron. Give it a photo. It gives back an image that looks like it was captured by a machine beyond human sight. Thermal camera. X-ray. Night vision. Kirlian plate.

It is not a filter. It does not paint new colors on top. It deletes the lighting of the original photo, then rebuilds the image from what things are: skin, hair, metal, water, leaves, not from how they were lit.

A photo records light from outside. A thermogram records heat from inside. Heron makes the second one. The name is from Heron of Alexandria. He built the first steam engine, and instruments to measure what no eye could. He also built machines just to make people wonder.

The noise, the colour, the chaos: Billions of people surrendering to the same obsession at the same time. And here I am, quietly building images from thousands of tiny fragments, one by one, asking myself: why does this feel exactly the same? #FIFAWorldCup

View full thread

That’s exactly why I built Shapearator: a local-first tool that identifies shapes inside a PNG or SVG sheet, extracts them one by one, and organizes them into separate files with metadata.

It uses computer vision, CV geometry, and export normalization to detect isolated forms, place them on a shared canvas, and generate structured outputs in formats like PNG, JPG, TIFF, and SVG.

In 1966, Bell Labs produced a portrait from electronic circuit symbols. Viewed up close, it was pure abstraction. Viewed from a distance, a human figure emerged.

View full thread

Nature continuously invites us to zoom in and out, shifting between fragments and wholes, details and patterns, pixels and narratives. For the past 3 decades, I have explored that territory with a little help from science, math, and technology.

In 1966, Bell Labs produced a portrait from electronic circuit symbols. Viewed up close, it was pure abstraction. Viewed from a distance, a human figure emerged.

The work was controversial enough to disappear quickly, but the problem it revealed would endure: how can meaning arise from the accumulation of discrete, locally insignificant parts?

I made a new open-source app called Hipparchus. It creates clean vector maps from OpenStreetMap data and exports them as layered SVG files. That means roads, buildings, water, parks, labels, railways, coastlines, and other elements stay separate and editable.

It uses local, open-source vision-language, computer vision, and language models to segment content, refine masks, and produce structured outputs for design and production pipelines.

For more than two years now, I’ve been experimenting with building software tools with AI assistants. People often call it #vibecoding. Personally, I never liked the term. Coding is only a language. The real goal is building systems that do something useful. tsevis.com/vibecoding-o...

tsevis.comCharis Tsevis - Beyond “Vibe Coding”Visual designer living on the sunny side of the planet. Photomosaics, computergraphics, neofuturism, digital art and graphic design

I've officially upgraded to the Qwen Coding Plan via Alibaba Cloud. While the initial setup required a bit more effort than anticipated, the value proposition is undeniable: access to 8 models for just $10/month. #AICoding