Case study · Draft

Impossible Frame

From an M.C. Escher drawing book on the kitchen table to a live WebGL plate archive on the web.

Jan BockJB Labs20269 min readLiving draft
Fig. 01a — The Impossible Frame, solid beams. Live geometry, not a baked plate.

A book on the table

As a kid I got an M.C. Escher drawing book from my parents. That was enough. I spent hours copying the impossible staircases, the interlocking animals, the rooms that refuse to decide which wall is in front. Penrose triangles. Optical tricks that look honest until you try to rebuild them.

I liked the patience of it. Pencil on paper. Lines that only work if every joint is true. If one corner is lazy, the whole paradox collapses into three ordinary beams.

New tools, same itch

Years later the itch is still there, but the tools changed. Shaders. WebGL. Real-time geometry. Communities like the Neue Academy that make it normal to ship experiments in public and talk about craft without pretending everything is a product launch.

So I gave it a shot. Not to recreate Escher as a poster. To ask a harder question: can an impossible object stay impossible when you move the camera, drag a slider, and export a frame?

If one corner is lazy, the whole paradox collapses into three ordinary beams.

Lovable, then the local loop

I started Impossible Frame in Lovable. Fast scaffolding, real routes, something on screen quickly. Then I plugged Cursor and Claude Code into the GitHub sync. That let me keep building locally, push to the connected branch, and save Lovable credits for the moments that actually need the hosted editor.

The workflow is deliberately hybrid: generate momentum in the cloud when it helps, refine geometry, shaders, and chrome in a proper repo when precision matters. This case study itself is still a draft. The studio is too.

Math first, graphite second

Most “Escher on the web” pieces are images, videos, or CSS tricks that fail the moment you orbit. Impossible Frame treats specimens as real Three.js geometry with a live stipple shader: stipple, hatch, grain, contour, all responding to light and pointer.

The Penrose collapse only holds along the isometric view. We use a near-orthographic perspective so the plate still reads flat while depth cues survive. Beams are merged and welded so the crease detector inks real silhouettes instead of seams. Half a beam width of tip offset decides whether the paradox stands or falls.

Bruno Ernst’s Zauberspiegel is the construction desk copy. Waterfall is three linked U-channel tribars plus a mill wheel under the free tip — Ernst’s perpetuum mobile after Abb. 190. Belvedere is the Spukhaus cuboid lie: decks whose long axes disagree, two honest corner columns, six crossed middle columns, and a ladder that is inside below and outside above. Tessellation borrows the Alhambra lesson: concrete interlocking motifs, not plain height-alternating boxes.

Page notes from the Zauberspiegel

Bruno Ernst, Der Zauberspiegel des Maurits Cornelis Escher (Berlin: Taco, 1986). Citations below come from a second OCR pass (PyMuPDF text layer + Tesseract on rendered pages 92–93, 94–95, 42–43).

Belvedere / Spukhaus (pp. 92–93, Abb. 181, 182a–c, 183, 185, 186, 187): Eight columns join the floors: only the extreme left and right are honest (edges AD and BC in Abb. 182a). The other six connect front to back and must run diagonally through the middle space. The ladder is straight yet stands inside below and leans on the outside wall above (Abb. 183). Mid-ladder you cannot say whether you are in or out.

Wasserfall (pp. 94–95, Abb. 188, 189, 190, 191, 192–199, 200): Escher links three such tribars (Abb. 190) into Wasserfall (1961). Early sketches were huge building complexes; falling water made the absurdity readable. Reading left-top: water falls, turns a mill wheel (Rad), then runs through a brick channel that appears to descend continuously away from us — until the farthest lowest point coincides with the nearest highest, a perpetuum mobile.

Flächenaufteilung (pp. 42–43): He kept concrete motifs (birds, fish, reptiles) where Islamic mosaic stayed abstract — recognizability is the engine of his Flächenaufteilung. Reptiles and related cycles use interlocking shapes whose head/notch pairs lock under 180° rotation — the motif language behind our tessellation plate.

OCR limits: The PDF is a page scan: embedded text is incomplete and OCR (eng tessdata) mangles umlauts. Diagram pages (esp. Abb. 182–190, 195–200) are image-heavy; captions OCR better than line art. deu.traineddata was not installed; German OCR accuracy is limited to eng fallback.

Fig. W — Three-tribar aqueduct + mill. After Ernst pp. 94–95, Abb. 190 / Wasserfall 1961.
Fig. B — Spukhaus column swap. After Ernst pp. 92–93, Abb. 182–183.
Fig. T — Notched reptile paving. After Ernst pp. 42–43, Alhambra → Reptiles.

A plate studio, not a demo reel

The startpage is a draggable plate archive. Open a specimen to tune. Compare constructions. Drop into The Lab when you want the full instrument. CSS illusions sit beside WebGL plates in the same catalog, because some paradoxes are optical and some are spatial.

The look is graphite on paper on purpose: architectural plate meets lab notebook. Brand-first Impossible Frame. Contact and CTAs back to jan-bock.com. Credits to the gallery patterns that sparked the chrome.

What this is

A childhood habit meeting a contemporary toolchain. An experiment in exact math transferred to the browser. A draft I am still drawing, now with shaders instead of only graphite.

If you want to talk craft, illusions on the web, or how this kind of studio work lands in a brand surface, visit jan-bock.com or write hello@jan-bock.com.