I Miss the Render
Ibrahim Fawakherji — ArchUp
A few days ago I opened my personal Gmail account after the message everyone dreads appeared on the screen. Storage at 98 percent. The familiar cleanup ritual began. Deleting files, clearing attachments, scrolling back through years of correspondence. Then I started opening the largest folders, moving backward through time. 2015. 2014. 2013.
And there, inside hundreds of old threads, I found something I was not looking for.
I found the render.
Not the images themselves. The entire world that existed around them.
Long correspondence chains with visualization studios in three different time zones. Revision lists that ran to forty items. Reference folders organized by material, by lighting condition, by camera angle. Virtual meetings that lasted two hours to discuss the position of a single shadow. One image in those archives cost more than 1,500 dollars. A single frame. Weeks of back and forth for one deliverable that the client would glance at for thirty seconds before asking for a different sky.
Today anyone with a reasonable prompt and ninety seconds can generate fifty versions of that same image. But in that period, the image was a project unto itself. And something about finding those old threads made me sit still for longer than I expected.
The architect has always been among the most technically layered professionals in any creative field. The skill stack was not shallow.
You began with hand drawing. Then AutoCAD. Then three-dimensional modeling. Then material preparation. Then lighting setup. Then rendering. Then post-production. Then the final presentation board.
Each stage was its own discipline. Each demanded time that could not be compressed without consequences.
What we were actually doing, underneath all the technical labor, was trying to understand the project before it was built. We were asking the rendering to answer questions the drawings could not. How would the sun fall on that wall at four in the afternoon in December? How would the glass read against the stone when both were wet? How would a person feel standing in that corner specifically, not the corner three meters to the left?
The render was not a marketing image. It was a thinking tool. The hours spent building it were hours spent interrogating the design.
And then we pressed the button and waited.
Those of us who worked through the V-Ray and Mental Ray era remember the particular quality of that wait.
You spent hours calibrating materials. The concrete mix. The glass reflectivity. The stone roughness value. The refraction index. The shadow softness. Every parameter required testing and the testing required rendering small samples, evaluating them, adjusting by increments, and rendering again.
Then came the moment the whole team recognized.
The render button.
It was, in its way, the closest thing in daily architectural practice to launching something. Months of work compressed into a single click, after which the machine took over and you simply waited to see whether the decisions you had made were right.
Some projects rendered in thirty minutes. Others ran through the night. I remember one graduation project where the machine ran for more than twenty-four hours. We would leave it working and return periodically to check progress, the way you check on something fragile and important. The image built itself line by line down the screen, and each pass revealed a little more of whether the thing you had imagined actually held together in light.
I kept one photograph from those years. My university computer, running a cooling system that had been modified to handle the thermal load of sustained rendering. The machine looked less like a workstation and more like something that needed maintenance. Tubes, fans, the particular hum of hardware being asked to work beyond its comfortable range.
That photograph is a document of a specific moment in architectural production. A moment when the gap between what you could imagine and what you could show was bridged by time, heat, and patience.
What came after the render was, in many ways, more interesting than the render itself.
The pass layers.
Reflection. Refraction. Material ID. Render ID. Shadow Pass. Ambient Occlusion. Sky. Lighting. Depth. Dozens of separate outputs that were assembled inside Photoshop into the final image. The process was closer to printmaking or darkroom photography than to anything we would now call digital production. Each layer carried specific information. The skill was in understanding what each layer contributed and how to combine them without losing the coherence of the whole.
It was craft work. Slow, specific, cumulative craft work.
And that is probably why I feel something when I scroll through those old files. Not nostalgia for worse results. Nostalgia for a longer road.
The economics of that era were substantial and are now almost entirely gone.
Architectural visualization was a serious industry. Studios in Eastern Europe, Southeast Asia, and the Gulf employed skilled teams whose sole function was translating architectural intent into photorealistic images. A mid-range exterior render from a reputable studio cost between 800 and 2,000 dollars. A full project package, covering key views, interior perspectives, and aerial shots, could run to 15,000 or 20,000 dollars for a significant commission. Larger firms maintained in-house visualization departments with dedicated hardware, licensed software, and specialists who spent their entire careers developing expertise in light simulation and material behavior.
That industry is being restructured at a speed that has no precedent in any previous software transition the profession has experienced.
The AI image generation market, valued at approximately 300 million dollars in 2022, is projected to exceed 900 million dollars by 2028 according to recent market analyses. The visualization studios that once charged per image are now being asked to justify their existence against tools that produce comparable outputs in seconds at marginal cost. Some are pivoting toward animation, toward technical accuracy, toward the kinds of outputs that generative AI still cannot reliably produce. Others are contracting. The workforce that built careers around V-Ray expertise is facing a restructuring that is not gradual.
This is the honest version of what is happening. The render economy is not declining. It is being rebuilt from different foundations, and the transition is moving faster than the people inside it can comfortably absorb.
A few days ago I was testing a current AI image generation tool against a reasonably complex architectural massing. The form involved torsion, cantilevers, and some specific geometric relationships that would require structural resolution in any real project.
The first output was striking. Genuinely impressive at the scale of a thumbnail.
Then I looked more carefully.
Details that were visually plausible but physically impossible. Structural elements that implied loads they could not carry. Connections between components that had no logic behind them. A building that looked like architecture but was not making the arguments that architecture is required to make.
The AI had generated the appearance of resolution without the resolution itself. It had learned what buildings look like without learning what buildings do. The image was confident in the way that a hallucination is confident: completely, and without the self-awareness that would allow it to recognize its own errors.
This is the current condition of generative AI in architectural visualization, and it is worth being precise about it. These tools are extraordinary at producing spatial atmosphere. They are genuinely useful for early concept communication, for mood, for massing studies, for presenting options to clients who need to feel the direction before the direction is fixed. They are not yet reliable for communicating technical intent, for representing constructable geometry, or for the kinds of claims that a serious architectural image is supposed to make about how a building will actually stand and perform.
The hallucination problem is not a minor limitation. It is a structural condition of how these models generate images, and it matters enormously in a discipline where the image is supposed to be a commitment, not a speculation.
I am not writing this as a defense of the past.
The tools that are replacing the V-Ray workflow are genuinely powerful and they will reshape the profession in ways that most practices are only beginning to understand. The architect who refuses to engage with them is making the same mistake as the architect who refused to engage with CAD in the early nineties.
But I am writing it as documentation.
For the generation that lived through the render era, and for the generation that never will.
There was a period in architectural practice when the quality of an image was measured in part by the hours it took to produce. When the machine was a collaborator that needed to be fed carefully and waited on patiently. When pressing render was a small act of faith, because you had done everything you could and now the calculation would tell you whether your decisions held.
That friction produced something. Not better images, necessarily. But a particular kind of attention to the decisions that preceded the image. A discipline of thinking that came from knowing the image would cost you time you could not get back.
When I scroll through those old email threads now, I am not mourning the workflow. I am recognizing something about what the workflow asked of the people who used it.
It asked them to be certain before they committed.
Because pressing render on a guess was an expensive way to discover you were wrong.
✦ ArchUp Editorial Insight
What this essay mourns as the disappearance of a workflow is more precisely the dismantling of a labor-intensive verification system that embedded design accountability into the production process itself — a system whose friction was not inefficiency but epistemology. The 1,500-dollar render was expensive not because visualization studios were overpriced, but because the hours required to simulate light behavior at the material level forced the architect into a sustained interrogation of decisions that cheaper tools allow to remain unresolved indefinitely. The generative AI image does not merely accelerate this process; it structurally decouples visual plausibility from physical coherence, producing outputs that are, as the article correctly identifies, confident without being correct — a condition that mirrors precisely what the compressed construction pressures examined in The Taptab Illusion produced in concrete: the appearance of completion masking the absence of the process that completion requires. The visualization economy that is being dismantled was not merely a service industry; it was, in its slower and more expensive form, a distributed quality-control mechanism for design thinking, and what replaces it — images generated in ninety seconds for near-zero marginal cost — transfers the accountability gap downstream, onto the contractor who must build what the hallucination promised, and onto the occupant who must inhabit the difference between the two.
