tenobrus/graphglyph

Graphglyph is a Python tool that hides text inside a mathematical graph image. You give it a string of text, and it produces a blue-and-orange diagram of dots connected by lines. The diagram looks like abstract geometric art, but the original text is fully recoverable from the image file. It works in reverse too: feed the image back to the tool and get the original text out. The hiding method works by varying the visual weight of edges in the graph. The graph is built from a specific family of points defined by a mathematical formula (based on research from an OpenAI paper on unit-distance graphs, where every connected pair of points is exactly distance 1 apart in a plane). The tool encodes your text as a series of four-bit values and distributes them through the graph by making one edge in each candidate pair slightly stronger or thicker than the other. A random seed derived from the text also shifts how the graph is structured, so two different phrases produce visibly different-looking images. The tool has three modes for generating the point set that underlies the graph. The default mode produces images that vary noticeably between different inputs. The other two modes follow specific bounded-norm variants from the mathematical paper, producing graphs with different densities and shapes. Output can be SVG (scalable vector format, fully decodable), JSON (data format, also decodable), or PNG (a visual preview that cannot be decoded back to text). Colors in the image are for display only and do not affect decoding. You can customize dot colors, edge colors, and background color. It is a single-script command-line tool with no notable dependencies beyond Python. The README notes that all code was written by GPT 5.5.