Charis Tsevis (@tsevis) | teapawt

Tweets
All Posts Highlights
Sort by
Most recent Popular
Tweets & Replies
Reposts
Media
Search

Pinned Tweet

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.

Algorithmic mosaic portrait of a young African woman in a colourful headwrap, looking straight at the viewer. Her face and wrap are built from hundreds of tiles of African textile patterns in terracotta, rose, indigo and gold. "Will You?" from African Bricks by Charis Tsevis.

ALT Algorithmic mosaic portrait of a young African woman in a colourful headwrap, looking straight at the viewer. Her face and wrap are built from hundreds of tiles of African textile patterns in terracotta, rose, indigo and gold. "Will You?" from African Bricks by Charis Tsevis.

Close-up of the same mosaic around one eye: every tile is a different printed pattern, zigzags, dots, diamonds and stripes in browns, pinks and terracotta, separated by thin dark lines.

ALT Close-up of the same mosaic around one eye: every tile is a different printed pattern, zigzags, dots, diamonds and stripes in browns, pinks and terracotta, separated by thin dark lines.

2

1

152

A thermal camera does not see light. It sees heat. So Heron does not tint your photo orange. It estimates what is in the frame (skin, hair, fabric, metal, glass), removes the photograph's lighting, and re-synthesizes the image from what things are. The coloured stage lights in the source are gone. A gradient-map filter cannot do that. Open source. Artistic simulation, not real thermography or medical imaging: github.com/tsevis/heron

Left: a portrait of a young woman under multicoloured stage lighting. Right: Heron's thermograph of it, in glowing yellow, orange and purple, with the coloured light removed.

ALT Left: a portrait of a young woman under multicoloured stage lighting. Right: Heron's thermograph of it, in glowing yellow, orange and purple, with the coloured light removed.

26

One face, three ways to assemble it from words. The man doesn't change. His mood does.

Three typographic mosaic portraits of the same man with blue eyes, looking up and aside. The left is soft and blended, the middle sharper with larger letters, the right built from tight, busy pieces.

ALT Three typographic mosaic portraits of the same man with blue eyes, looking up and aside. The left is soft and blended, the middle sharper with larger letters, the right built from tight, busy pieces.

51

Charis Tsevis retweeted

@QwenAIFansNov 1

⚡️ Herbert just beat Apple’s MLX on Qwen3. Zero MLX code. On a Mac Studio M3 Ultra: up to 31% faster than mlx-lm across six Qwen3 models in bf16. Long prompts shrink the lead. Still wild for local Mac. via Philippe Anel 👇 philippe-anel.fr/en/blog/202…

2

11

1,161

Handoff. Steve Jobs, Tim Cook and John Ternus, made of 585 macOS icons, none recoloured. And one more thing: Steve Wozniak, where it all began. tsevis.com/portfolio/handoff

Steve Jobs mosaic portrait made of macOS icons, filling the frame, by Charis Tsevis

ALT Steve Jobs mosaic portrait made of macOS icons, filling the frame, by Charis Tsevis

Tim Cook mosaic portrait made of macOS icons, blue glasses, on white, by Charis Tsevis

ALT Tim Cook mosaic portrait made of macOS icons, blue glasses, on white, by Charis Tsevis

John Ternus smiling, mosaic portrait made of macOS icons, by Charis Tsevis

ALT John Ternus smiling, mosaic portrait made of macOS icons, by Charis Tsevis

Steve Wozniak smiling, mosaic portrait made of macOS icons, by Charis Tsevis

ALT Steve Wozniak smiling, mosaic portrait made of macOS icons, by Charis Tsevis

1

269

Philon, a macOS app I made to turn PDFs and scans into Markdown, with the evidence for every line kept beside it. Where it is unsure, it says so instead of guessing. For anything you need to cite. github.com/tsevis/philon

Philon's workspace in three panels: on the left the source page of a scientific PDF, in the middle the same page converted to clean readable text, on the right the evidence for the selected block (method, 98% confidence) and orange warnings saying the reading order is ambiguous and was kept for review.

ALT Philon's workspace in three panels: on the left the source page of a scientific PDF, in the middle the same page converted to clean readable text, on the right the evidence for the selected block (method, 98% confidence) and orange warnings saying the reading order is ambiguous and was kept for review.

53

WIP study: a clip rebuilt from basketballs, 8.9 seconds, 213 frames. The vision part: SAM 3, prompted with "basketball", on every second frame, interpolated in between. It found the ball in 106 of 107 frames. Its last 0.75 s become a wake of bigger balls.

1

2

261

I did not cut the player out with a matte in these clips. The figure is whatever small balls the picture's edges produce; the top 20% and bottom 15% keep only those and the wake. In the earlier stills I did use a matte (BiRefNet). Not sure yet which one reads better.

Close-up of black-and-white basketball tiles packed into the shape of a head and torso on white, with a few larger balls at the upper right.

ALT Close-up of black-and-white basketball tiles packed into the shape of a head and torso on white, with a few larger balls at the upper right.

1

1

28

Open problem: a 0.3 grey-level change in a still background changed 22% of the mosaic's pixels in an early test. Holding the circles still helped, not enough. How do you keep a tile choice stable across frames? Source image: Paolo Guidobono. tsevis.com/blog/every-tile-i…

Every Tile Is a Basketball: An Algorithmic Mosaic Study — Tsevis Studio

A study in algorithmic mosaics: a basketball player rebuilt from black-and-white basketballs, with short notes on the computer vision that finds the ball, and references.

1

18

A mosaic that forgets the face. Abbott Thayer, the painter often called the father of camouflage, turns into pure camouflage in 14 seconds. Every frame is a real render: no cross-fade, no mask. One setting changes. The setting is Follow: how closely the strokes obey the portrait. At 1.0 the face is built from strokes. At 0 nothing of it is left. Length, width, bend and the palette move with it, from the photograph's tones to the colours of the earth. The vision is classical. k-means gives four tones. The photo becomes a lightness map. The strokes turn along the gradient of the blurred tone, so they follow the face's contours until Follow lets go. The dull bug: strokes were placed at random, so a small step in Follow re-rolled the pattern and the film flickered. Each stroke now has its own random numbers, from the seed alone. Work in progress. tsevis.com/blog/thayer-vanis…

3

224

Three approaches to one portrait: calm, precise, bold. None is "the right one". Which would you keep?

Three typographic mosaic portraits of the same man looking up and to the side, built from cut-out words and letters in warm skin tones against blue. The left is calm and softly blended, the middle is more precise in the eyes and mouth, the right is bolder with larger letters.

ALT Three typographic mosaic portraits of the same man looking up and to the side, built from cut-out words and letters in warm skin tones against blue. The left is calm and softly blended, the middle is more precise in the eyes and mouth, the right is bolder with larger letters.

2

2

112

Idea in progress: a mosaic whose tiles are words. The new tab takes the image of a page and cuts it into word and letter tiles with transparent backgrounds, so a manuscript becomes the tile library. First question: how small can a word get before it stops being a word and becomes texture?

63

📚📖👨‍🏫

@ajitcodesOct 31

Andrej Karpathy dropped a 2-hour masterclass on everything that he knows It covers: - Agents - ⁠Harnesses - ⁠Loops - ⁠Graphs This is a must watch. Bookmark this post

57

One white ball. Then three. Then hundreds. Then a player. 1,885 basketballs, placed by Mozaix's Circle Packing engine: small where he moves, big where he doesn't. In Blender each ball bounces to the height of its own point in the image. How many seconds before you saw him? 👇

1

91

Kitaoka's illusions work on the visual system, not on the picture. I wrote a small generator for his diagonal-bevel pattern (SVG out, plain Python, no dependencies), and a studio that makes the same patterns as mosaic tiles: 10 shapes, 12 colour presets. The idea: a mosaic whose pieces build the picture and the illusion at once.

A grid of slate-blue and orange squares, each framed by light and dark diagonal bevels, generated in the style of Akiyoshi Kitaoka's optical illusions.

ALT A grid of slate-blue and orange squares, each framed by light and dark diagonal bevels, generated in the style of Akiyoshi Kitaoka's optical illusions.

1

45

"So long have I gazed upon beauty that my sight is filled with it." A typographic mosaic portrait of Cavafy, written word by word. The voice is Elli Lambeti, reciting "Έτσι πολύ ατένισα" (1917), and the music is mine. #Cavafy #Καβάφης #poetry #codingpoetry #algoritmicmosaic

1

5

293

Την εμορφιά έτσι πολύ ατένισα, που πλήρης είναι αυτής η όρασίς μου. Γραμμές του σώματος. Κόκκινα χείλη. Μέλη ηδονικά. Μαλλιά σαν από αγάλματα ελληνικά παρμένα· πάντα έμορφα, κι αχτένιστα σαν είναι, και πέφτουν, λίγο, επάνω στ' άσπρα μέτωπα. Πρόσωπα της αγάπης, όπως τα ήθελεν η ποίησίς μου … μες στες νύχτες της νεότητός μου, μέσα στες νύχτες μου, κρυφά, συναντημένα …

1

43

So long have I gazed upon beauty that my sight is filled with it. Lines of the body. Red lips. Limbs made for pleasure. Hair as if taken from Greek statues, always lovely, even when left uncombed, falling a little over white foreheads. Faces of love, the way my poetry wished them … in the nights of my youth, in my nights, met in secret … — C.P. Cavafy, 1917

30

Leon Harmon reduced a portrait of Lincoln to a coarse grid, around 16 x 16 blocks. You can still recognise him. Lay the wrong pattern over it, at the wrong spatial frequency, and the face is gone. Dalí saw the image and turned it into a painting in 1976. That is a mosaicist's daily problem: the detail inside the tiles can live with the face, or fight it. tsevis.com/blog/history-of-m…

Two coarse, textured portraits of Abraham Lincoln from Leon Harmon's face-recognition research: the same face under different overlaid patterns, one more recognisable than the other.

ALT Two coarse, textured portraits of Abraham Lincoln from Leon Harmon's face-recognition research: the same face under different overlaid patterns, one more recognisable than the other.

1

65

The more the internet fills with digital rubbish, AI slop and content made for agents, the more we need journalism to find its way again, and information and culture curated at a human scale. This is what newspapers and magazines offer at their best: professionals bringing knowledge, expertise and a point of view to the task of making sense of the world. Finite editions we can spend time with. Something to chew over quietly, digest and turn into knowledge or creativity, instead of consuming compulsively. LIFE’s return to print is not the first sign of this renewed appetite. Excellent publications have been emerging for some time. But @LIFE is a symbol! I wish it every success, and the same for an entire profession that needs to see itself again as a source of hope, with the judgment and independence to earn our trust. #Lifemagazine

Image from Charis Tsevis (@tsevis)

57

This is so nice!

Johannes Mutter@JohannesMutterOct 28

For 15+ years I’ve collected the interfaces that made me stop and ask “wait, how does this work?” Interface Lineage is now open interfacelineage.com Learn the patterns and principles that turn “make no mistakes” into clear instructions for AI. 500+ interfaces, free to read.

69