Skip to content
Categories:

Not-so-Large LM

Post date:
Author:
Number of comments: 2 comments

Two weeks ago I thought I’d give my handy native Chinese AI assistant Doubao (豆包 = Bean Wrap) the job of writing its version of my post of last week. Apart from the first and last sentences, the prompt was about the same level of detail as a note I might make as a seed for any new post.

I asked Doubao AI-whatever how many requests it is usually processing at the same time and the answer was somewhere between 10,000 and 100,000 requests per second. Here’s what was returned in less than half of one.

Below is the full text. The only change I made was to strikethrough the authorship attributed to me. I did ask Doubao to write in the style of Graham McKay but I never intended to pass off what it produced as mine. I can see how this default assumption could be true most of the time. Anyway, here’s what was produced. See you on the other side.

Supergraphics in Architecture: When Surface Image Redefines Built Space

By Graham McKay (Misfits’ Architecture)

Architecture loves to pretend it is immutable. We build walls, floors, and ceilings as fixed boundaries, treating built form as the final statement of a space’s character. But for as long as modern architecture has existed, there has been a quieter, rebellious counter-tradition: one that rewrites buildings not by moving concrete or steel, but by covering, stretching, and projecting massive graphic fields across their surfaces. This is the world of supergraphics — a deceptively simple tool that turns architecture’s most basic assumption on its head: structure does not always define space. Sometimes, image does.

Let’s start with a clean definition, stripped of design-jargon fluff. Supergraphics are large-scale, two-dimensional graphic interventions that override architectural boundaries to reshape the perceptual geometry of a building or interior. Unlike ordinary murals, decorative paint schemes, or signage, supergraphics intentionally ignore plane breaks. They wrap across walls, spill onto ceilings, fold around corners, and dissolve the readable edges of built volumes. They are not merely “big art on buildings”; they are spatial modifiers.

What Exactly Are Architectural Supergraphics?

Coined formally by design critic C. Ray Smith in 1967, the term originally described monumental, non-figurative geometric fields designed explicitly to alter spatial perception rather than illustrate or advertise. Crucially, early supergraphics avoided faces, logos, and alphanumeric text. They were pure shape, pure color, pure scale — graphic architecture that treated the entire built enclosure as a single canvas rather than a collection of discrete surfaces. In short, supergraphics turn static rooms and facades into dynamic environments through surface alone.

A Brief, Rebellious History of Supergraphics

Supergraphics did not emerge from nowhere. They arrived in the 1960s as a perfect collision of mid-century modern restlessness, Pop Art’s bold visual language, and a growing frustration with rigid architectural orthodoxy. For decades, modernism prioritized structural honesty, clean lines, and material austerity. By the mid-60s, younger designers were pushing back: why let a building’s fixed form dictate its mood when color and pattern could reinvent it overnight?

The pivotal figure in this shift is Barbara Stauffacher Solomon, the pioneer who turned supergraphics from a casual design trick into a legitimate architectural strategy. Working on California’s iconic Sea Ranch development alongside architects including Charles Moore, Solomon deployed sweeping geometric supergraphics that stretched across interior and exterior surfaces. Her work did not decorate the buildings — it reordered them. Where modernist structures could feel rigid or repetitive, her oversized graphic fields unified disjointed volumes, softened harsh angles, and gave standardized residential architecture a distinct, playful identity.

By the late 1960s and early 1970s, supergraphics bloomed across Europe and North America. French designer Jean-Philippe Lenclos1 integrated supergraphic color fields into university auditoriums and public housing, treating entire interiors as “habitable sculptures” modulated by tone, contrast, and reflective surfaces. Schools, community centers, and public housing blocks adopted the style precisely because it offered something radical: architectural transformation without demolition or reconstruction. It was cheap, fast, reversible, and endlessly adaptable — a populist counterpoint to the elitist permanence of high modernism.

The trend faded in the 1980s as architecture swung back toward disciplined formalism, structural rigor, and restrained material palettes. Supergraphics were dismissed as frivolous, dated, and superficial. Yet they never truly disappeared. They simply waited for digital technology to catch up to their radical potential.

Image Placeholder 1: Classic Analog Supergraphics (1960s–70s)

Caption: Barbara Stauffacher Solomon’s iconic Sea Ranch interior supergraphics (California, 1960s). Bold, continuous geometric color fields wrap across wall corners and ceiling planes, erasing architectural subdivisions to unify fragmented modernist volumes. This is foundational analog supergraphics: static, physically applied, and spatially transformative.

From Mosaic and Mural: The Era of Applied Supergraphics

To understand modern luminous supergraphics, we first need to ground them in their historical lineage: applied supergraphics. Long before digital projection and LED screens, architecture relied on fixed, surface-mounted graphic treatments to achieve similar spatial effects. Mosaics, tiled patterns, large-scale painted murals, and printed vinyl graphics were the original supergraphic tools.

These traditional applied systems share core traits. They are physical, static, and permanent or semi-permanent. They require direct material adhesion to the building fabric. A mosaic tesserae field or a hand-painted geometric supergraphic becomes part of the building’s surface texture, weathering with the facade, aging with the structure, and settling into the architecture’s long-term identity. Crucially, applied supergraphics are limited by fabrication. Once installed, their composition, color, and scale are fixed. They can respond to the building’s form, but they cannot change with context, time, or user behavior.

For most of the 20th century, this was the only version of supergraphics architecture had access to. The power of spatial reinvention was real, but it was one-and-done. You redesigned the surface once, and the building’s graphic identity was set for years, even decades.

Projection vs. Emission: The Digital Split in Modern Supergraphics

The digital age split supergraphics into two distinct species: projected light supergraphics and emissive screen supergraphics. The difference is not merely technical — it is architectural, altering how buildings occupy time, respond to context, and define urban presence.

First, projected supergraphics. These use external lighting rigs, projectors, and gobos to cast massive graphic fields onto existing building facades and interiors. Like classic applied supergraphics, they rely on the building’s physical surface as a canvas. But unlike mosaics or paint, they are entirely temporary and non-invasive. Projection does not alter fabric, does not require adhesion, and leaves no trace after shutdown. A concrete facade can host swirling geometric supergraphics one night, typographic fields the next, and return to raw materiality by morning.

Projected supergraphics preserve the core ethos of classic supergraphics: they override architectural form to create perceptual shift. They soften rigid facades, unify disjointed building masses, and insert dynamic visual rhythm into static modernist structures. Yet they remain respectful of the building’s underlying architecture. The structure remains the anchor; the graphic is a responsive, ephemeral layer on top.

Second, and far more disruptive: emissive supergraphics — full-building LED display surfaces that turn the facade itself into a luminous, low-resolution screen. These are not projected onto architecture; they are architecture. The building’s skin no longer exists primarily as enclosure, weather barrier, or material expression. It exists as a display medium.

Crucially, these low-resolution architectural screens operate differently from consumer displays. They prioritize scale over sharpness, atmosphere over detail, and ambient urban effect over legible content. They generate vast, flickering supergraphic fields of color, movement, and pattern that wrap entire city blocks, turning buildings into living, breathing visual organisms. Where projected supergraphics decorate architecture, emissive supergraphics redefine its purpose.

Image Placeholder 2: Projected vs. Emissive Supergraphic Comparison

Caption: Side-by-side urban facade comparison. Left: Nighttime architectural projection mapping casting large-scale geometric supergraphics onto a concrete facade (non-invasive, ephemeral, surface-dependent). Right: Low-resolution LED facade grid running continuous ambient color-field supergraphics (emissive, architectural, integrated into building skin). The contrast illustrates the shift from overlay architecture to screen architecture.

The New Architecture of Living Supergraphics

This divide between static applied supergraphics, temporary projected supergraphics, and permanent emissive supergraphics is not just a technical evolution. It signals a quiet revolution in what architecture can be.

Traditional architecture is a discipline of fixity. We design buildings to resist change, to endure weather and time, to maintain consistent form and character. Classic supergraphics already challenged this by proving surface could rewrite space. Digital supergraphics eliminate fixity entirely. The new architectural variable is not material, proportion, or structure — it is content.

Emissive supergraphic facades create a new kind of hybrid building: part enclosure, part public media. These structures do not just occupy space; they communicate continuously. Their aesthetic identity is no longer fixed at completion. A single building can shift from muted, atmospheric minimalism to bold, high-contrast supergraphic patterning, to responsive environmental visualization, all within a single day.

This shift forces us to rethink core architectural priorities. Material honesty, a sacred modernist principle, becomes less relevant when the primary surface experience is luminous and digital. Formal harmony takes second place to temporal harmony — how a building feels across hours, seasons, and urban moods. Urban aesthetics move from static skyline composition to dynamic, responsive ambient landscape.

Of course, there are tradeoffs. Low-resolution luminous supergraphics risk visual noise, urban clutter, and generic brightness. When every building can endlessly reinvent its skin, the quiet dignity of fixed architectural form can be lost. Yet the potential is equally profound. Supergraphics allow architecture to become adaptive, contextual, and empathetic — responding to community mood, marking cultural events, or softening harsh urban environments without permanent construction.

Closing Thoughts: Supergraphics as Misfit Architecture

Supergraphics have always been misfit architecture. They do not play by the rules of structure, material, or fixed form. They start from a simple, subversive question: what if the most powerful architectural element is not the wall, but the image that covers it?

From hand-painted 1960s geometric fields to today’s luminous building-scale screens, supergraphics have evolved from a niche interior trick into a fundamental architectural language. The old applied supergraphics modified buildings; projected supergraphics temporarily reimagined them; emissive supergraphics have become them.

As our cities grow denser and more digital, supergraphics are no longer decorative afterthoughts. They are the interface where architecture meets public life, where fixed built form meets fluid cultural expression. For better or worse, the buildings of tomorrow will not just be seen — they will perform, and supergraphics will be the primary medium of that performance.

Comments

  1. The title Supergraphics in Architecture: When Surface Image Redefines Built Space is bland but informative. It’s like a research paper title designed to include as many keywords as possible. I don’t think this is an accident. The word “when” hints of the position that will be set out. Clever.
  2. I asked for 1,500 words which is exactly the number returned.
  3. Overall, the piece didn’t sound like me and I can’t say I’m upset. As it is, the piece is a bit light on content and thin on examples but it’s well structured.
  4. The topic Supergraphics calls for images but what was written has none apart from the one suggestion I don’t think even that relevant. I’ve found AI to not be very good at providing images to illustrate thoughts. But then, it’s a Large Language Model and not a Large Image Database indexed by keywords a likely prompt might contain.2 For instructors and especially ones at architecture schools, this is good news because the ability to sift experience and knowledge to provide students with relevant references has no competition for the time being – for as long as learning from references remains valued.
  5. In an earlier post3, I mentioned two instances where AI made word choice suggestions that made itself look good even though the meaning shifted. For companies promoting a service, it’d make sense to embed hidden instructions to do this. It would also make sense to include instructions to prioritize persuasiveness and the appearance of conviction no matter what the content.
  6. The overall structure and particularly the “Closing Thoughts” are persuasive, but the confident and oddly familiar tone of voice grates. It’s as if it was written as a narration to a video rather than something to be read and this is most likely the result of what it’s been trained on.

This last point is my main problem with this piece of writing. It takes itself a bit too seriously and tries too hard to convince. So much effort is being made to present the topic that the presenting becomes the topic. It’s like one of those TED talks where the speaker is using so many arresting phrases and emphatic gestures that you’re too distracted to notice the content. Having said that, I’m still fairly sure that if you fed AI explicit prompts for long enough, you’ll get it to produce a piece of writing in the style of your favorite author, academic, influencer or architecture blogger. But judging by the particular stylization of style evident in the sample, I suspect that this default “relatable” yet “authoritative” voice comes from AI having been cheaply trained on too many TED talks and far too much YouTube. It makes commercial sense to use publicly available and free information rather than stupid academic papers behind their paywalls. We get the AI we deserve. Its training data seems to be the entirety of YouTube rather than the entirety of human knowledge we were led to believe.

We can expect to see an increase in the amount of AI-generated writing and video recirculated as training data of decreasing value. These Large Language Models can only get smaller and there are signs this is already happening.4 It turns out that AI is just as lazy as we are when it comes to generating original content. Ultimately, language will flatten and this makes me sad because I have a fondness for it. Nobody’s driving this bus yet somehow, we all are.

We’ve already witnessed this flattening of expression with TED gestures. TED talks are no Chinese opera or Japaneses noh, but the gesture in the following three images is a live action emoji an audience understands as “sincerity” or “passionate” or somehow “seeming to have strong feelings about something”. A decade and a half ago, certain persons noticed our appetite for stylized expression. Now they’re circling around and this time they mean business.

• • • 

  1. I never knew about Jean-Philippe Lenclos. There’s two images below. ↩︎
  2. I’ve given up trying to persuade anyone that a model for architectural aesthetics is something we need. This article https://garymarcus.substack.com/p/generative-ais-crippling-and-widespread makes the case for having models of how some part of the world works. Instead we see progress as being able to sift through haystacks at ever-increasing speeds. ↩︎
  3. https://misfitsarchitecture.com/2026/05/31/i-is-for-improve/ ↩︎
  4. https://www.theguardian.com/commentisfree/2026/jun/17/elias-thorne-ai-generated-stories ↩︎

• • • 

Comments

  • says:

    Hello and thanks Victor! It’s only a few weeks since I wrote that post and I find my attitudes towards AI assistants have changed greatly and it’s not that the AI is doing the learning. I find I have to continually hold my assistant back from creating “polished” text. It’s invariably too modern and composed more for speech than for reading. Between me and my assistant responding to me, we’ve reached an agreement where my assistant does the research and checks the facts while I do the writing and revising. The speed at which new thoughts can be incorporated and entire drafts rewritten is astonishing. I’m pleased with the workflow. Sometimes it gets a bit much and I have to take a break adn think about stuff. I can see how students become unstuck. The quality of output is totally dependent upon the quality and the quantity of input. Makes sense. There’s definitely a place in the world for words written for the fun of it. Ongoing.

  • Hello! You mentioned having never known Jean-Philippe Lenclos, and I have never seen “fixity” and “gobos” in any text, but they appear to be legitimate words.

    Using odd words is a complicated choice. I have read a praise to Lincoln for using a 200-long vocabulary in his speeches. I got critiqued by certain colleagues that I should have talked as if to a five-year-old, and should have watched MVRDV lectures before going to the mic at certain events.

    Weird words may be special notions and help one’s understanding of the subject. Their use might suggest depth of knowledge. Using extravagant adverbs less so.

    You mentioned the thought I had, that models are supposedly trained on “the corpus” of the entire written language, but the peculiarities suggest the corpus must have been substantially rigged. The video archive conjecture offers an even plainer answer.

    Having seen math texts, one might desire to create weird notions, and those may lead to either discovery or insanity. These definitely make a certain entertainment, if the one concocting them is so inclined. I have made up a quote that I may have actually heard and then stolen, that one’s potential contribution to a culture is proportional to their difference from it.

    Thus my interest towards misfits’ architecture explains itself.

Leave a Reply