I started the layout at three in the afternoon, when the light in the studio comes through the polymer windows at the angle that makes everything look like it already belongs in a frame.
Twelve images. Fernanda had asked for twelve. The Vila Madalena gallery she runs — Espaço Aberto, which sounds grander than the converted shopfront it actually is — wanted twelve photographs from my editorial work, each accompanied by its process documentation. The idea was to show the audience what goes into making an image: the human decisions, the AI-assisted steps, the full chain of creation from concept to final file. "Transparency as exhibition," Fernanda had called it, which made me think of aquariums.
I printed the twelve photographs first. They looked like what they were: commercial fashion editorial, Brazilian light, color saturated the way my grandmother's darkroom prints used to be but achieved through different chemistry. AI-assisted color grading on seven of the twelve. AI background generation on two. AI compositing assistance on four. All of it disclosed, documented, timestamped. My entire career reduced to metadata.
I laid them out on the studio table and then I laid out the process timelines below each one.
The timelines were longer than the photographs were wide.
I had not expected this. When you document every step — every tool opened, every parameter adjusted, every moment a human hand overrode an AI suggestion, every moment it didn't — the record of making a photograph turns out to be larger than the photograph itself. Image number seven, a fashion editorial for a São Paulo streetwear brand, had forty-three documented steps. The final image was 4000 by 6000 pixels. The timeline, printed at the same font size as the image caption, was 1.2 meters long.
I stood there looking at it and thought: the timeline is the portrait.
That was when the concept changed.
I spent the next hour rearranging everything. The timelines moved up. The photographs moved down, printed smaller — sixty percent of their original size, mounted below their own creation records like footnotes to a text. The process documentation was the primary work. The photographs were illustrations of what the process had produced.
Fernanda arrived at five. She walked the length of the table, said nothing for two minutes — I counted, because I count things now; the spreadsheet has made me a person who counts — and then said: "The timelines are the portraits."
I nodded.
She said: "How many of the twelve are AI-assisted?"
I said: "All twelve have at least one AI-assisted step."
She said: "And the timelines show which steps?"
"Every step. Human, AI, and the moments between."
She looked at the longest timeline — image seven, the 1.2-meter streetwear editorial — and said: "This one is the best piece in the show."
I agreed, and then I added a thirteenth piece.
Row 149 of my booking spreadsheet. Printed at the same scale as the timelines, unlabeled, mounted on the far wall where you would see it last if you walked the gallery left to right. The row showed: date, client, project type, disclosure status ("full"), platform detection score (0.72 — just above the AI-generated threshold on Instagram, well within it on Behance), booking outcome ("paused pending certification"), estimated revenue lost (R$14,200), cumulative lost revenue since February (R$187,000).
I did not write a wall text for it. If you know what a booking spreadsheet looks like, you know what row 149 is. If you don't, it's just numbers in a grid. The gallery show is about process visibility. The cost of process visibility is also a process. It also deserves to be visible.
Jjang sat on the corner of the table and watched me arrange the thirteenth piece. She has opinions about art direction — her tail twitches when something is wrong — and she was still. I took this as approval.
Beatriz arrived at eight with her laptop under her arm and dark circles under her eyes.
I had made rice and a pork stew that my grandmother used to make — Korean-Brazilian comfort food, gochugaru and cumin, the kind of meal you make when someone needs to sit down and eat before they talk. Beatriz had been living on coffee and process documentation since the cease-and-desist letter arrived. Three letters now. The latest one didn't mention the NDA — it claimed her Selo de Processo codebase used "proprietary detection methodology concepts." She had published it on the Selo project page, highlighted, because that is who Beatriz is: if they want to sue her for making things visible, she will make the lawsuit visible too.
She ate half a bowl of rice before she looked at the laptop I had left open on the table, showing the gallery layout.
"What is this?"
"The show for Fernanda's gallery. Twelve photographs — well, thirteen now. The process timelines are the main work. The photographs are underneath."
She leaned forward. Scrolled slowly through the layout. Stopped at image seven — the 1.2-meter timeline. Zoomed in on the timestamped steps: 14:07 — opened Lightroom, imported RAW files. 14:12 — AI color grading applied (tool: MidGrade v3.1, preset: São Paulo Golden, modified +0.3 warmth). 14:14 — human override on skin tone correction, reverted AI suggestion. 14:15 — AI suggestion accepted for background color balance...
She read the whole thing. It took four minutes.
"You are building a courtroom exhibit," she said.
I shook my head. "I am building an exhibition."
"Same thing."
We ate in silence for a minute. The stew was better than I had expected — I had worried about the cumin ratio, but grandmother's recipe has never failed me. Jjang appeared from the bedroom, assessed the social situation, and chose Beatriz's lap. Beatriz stroked her without looking down.
"The detection system I built," Beatriz said, still looking at the layout on screen, "was supposed to make process visible. That was the pitch when I joined the company. 'We are building transparency tools.' But what it actually made visible was classification. The output was a number — 0.0 to 1.0 — and that number decided which tier you landed in. It didn't show anyone how you worked. It showed the platform how to sort you."
She paused. I waited.
"Your show does what the detection system was supposed to do."
I said nothing. I knew she was right, but I also knew the gallery show could not scale the way a detection system scales. Twelve timelines in a Vila Madalena shopfront reach maybe three hundred people. A detection algorithm reaches every image uploaded to a platform. The difference in scale is the difference between art and infrastructure.
Beatriz knew this too. She was already past it.
"I want to rebuild the Selo's technical architecture," she said. "Right now the Selo asks artists to publish their workflows as documentation. That's good. But the documentation is secondary — it sits next to the work, like a nutrition label. What if the timeline is the primary object? What if the Selo doesn't certify that your process was transparent — it certifies that your process is the thing people see first?"
I thought about Fernanda saying "the timelines are the portraits." I thought about the 1.2-meter record of image seven, which was more interesting than image seven.
"Like the gallery," I said.
"Like the gallery, but as infrastructure. Every Selo-certified work comes with a timeline as the primary file. The image or the text or the design is the attachment. The process is the work."
"The platforms won't index it that way."
"The platforms don't index the Selo at all. That's the whole problem. But if the Selo creates a standard where the process file IS the primary artifact, then anyone who encounters Selo-certified work encounters the process first. The detection system becomes irrelevant — you don't need an algorithm to guess how something was made if the creator is showing you, in detail, at the top of the file."
I looked at row 149 on the wall.
"Start with row 149," I said.
Beatriz looked at me. Looked at the unlabeled thirteenth piece. Looked back at me.
"What is it?"
"The cost. Timestamped. Disclosed. No wall text."
She understood. The process of making art transparent has a cost, and the cost is also a process, and both deserve to be visible. If the Selo is going to make the timeline the primary object, the timeline should include what the timeline costs.
She opened her laptop. The Selo project page was still showing the cease-and-desist letter, highlighted in yellow. 2,400 views, 47 shares, one unanswered email from a pro bono IP lawyer named Dr. Mendes.
"I should answer Dr. Mendes," she said.
"You should answer Dr. Mendes."
She started typing. Jjang settled deeper into her lap. I cleared the rice bowls and left the laptops.
Outside, São Paulo was doing what São Paulo does at nine in the evening — honking, breathing, sixteen million people in various states of disclosure and concealment. The three-tier sort was running on every platform, classifying every upload, assigning every image a number that determined its economic value. My number was 0.72. My ranking was 847. My cumulative lost revenue was R$187,000.
But in a gallery in Vila Madalena, the timeline was going to be the portrait. And in the Selo's next release, the timeline was going to be the file. And on the far wall, row 149 was going to sit unlabeled, the cost made visible alongside the process, which is the whole point and always has been.
I opened my booking spreadsheet. Row 150: tomorrow's shoot, a Brazilian jewelry brand, domestic market only, no EU labeling requirements. Detection score irrelevant. I would document the process anyway. I would make the timeline longer than the photograph.
I am a photographer who makes timelines now. That is what the sorting did to me. I am not sure it is the worst thing that has ever happened to my career. I am not sure it is not.
Jjang's tail twitched. I took this as acknowledgment.
Seo-jin's booking spreadsheet, row 149, sat on the gallery wall for six weeks. Forty-three people asked Fernanda what it was. Fernanda told them to look closer. Seven of the forty-three understood. The other thirty-six looked at the numbers and saw a grid.
Beatriz answered Dr. Mendes. The Selo's next technical release included a process-primary file format. The detection companies did not notice.
The three-tier sort continued.
Row 150 was documented. Row 151 was documented. The timelines kept getting longer.
The photographs, underneath, were still beautiful.