A camera cannot see the oil

Make a data-plate: one figure sized so it reads before the words do, sitting on a real photographed surface. One still image, 9:16 tall, briefed for a GPT Image 2 class model whose real ratio list is 1:1, 4:3, 3:4, 16:9, 9:16, 3:2 and 2:3. The discipline of this family is restraint. One number, one picture, nothing else competing. FACT (locked, build on it, do not re-research): researchers at the National Institutes of Health tested 4 widely used phone apps that estimate a meal from a photograph, against meals that had been prepared and measured precisely. The apps underestimated the meals by roughly 250 to 345 calories on average, about a third of what was actually there. Fat was the part they missed most, under by around 30 grams per meal, while carbohydrates were estimated far more consistently. High-fat dishes gave the apps the most trouble. The reason is the interesting part: fat that has soaked into food, or been absorbed during cooking, or pooled underneath it, is largely invisible in a photograph. A camera can see the shape and the colour of a meal. It cannot see what has gone inside it. Use these exactly and leave out any figure we have not locked. STATE OF THE EVIDENCE, and it must not be overstated: this was presented as a conference abstract at NUTRITION 2026 and has not been through full peer review. Do not draw it as settled science and do not invent an app, a brand or a sample size. TONE, and this matters: this is a piece about the limits of machine vision, not a piece about diets, weight or bodies. No people, no bathroom scales, no before and after, no guilt, no wagging finger. The subject is the thing the camera missed. NAMES: do not render any real app name, logo, icon or user interface in the artwork. GROUND (decide this, do not let the model default): a real photographed surface with history in it. Chipped diner formica, a waxed oilcloth, scratched steel, a cafe table with rings on it. Something a plate would actually be put down on. White, pale grey and navy are the defaults this brief FAILS on, and so is a clean white studio cyclorama, a wooden chopping board styled for social media, and a dark moody restaurant shot with a shallow depth of field. LIGHT: hard and slightly unflattering, overhead, the light a real meal is actually eaten under. DISTANCE: close, the meal filling most of the frame, near enough that the surface texture and the sheen of the food have real edge. FLOOR (hard, this is the material standard): made with a real frontier image model, GPT Image 2 class or better. Flat vector, hand drawn SVG or HTML, and screenshot collage all FAIL. Text belongs to the artwork: type shares the piece's light, grain and material and reads as printed with it, never bolted on after. Archives will not be considered. CRAFT STANDARD (read all of it before you start). Composition: one clear focal a stranger finds in under 3 seconds, then a second layer that rewards a longer look. Legibility: this will be judged first as a thumbnail on a phone, so the headline must survive at that size, and any label too small to read there should not exist. Palette: pick 2 or 3 inks and hold them. A piece with 8 colours reads as a template. Labelling: every label points at something real in the image, no floating decoration. Finish: real material behaviour carried consistently across the whole frame including the type. DIRECTION is yours. FOCAL: the invisible part made visible. The oil, the fat, the missing third, given a physical form the eye can measure against the meal itself. That is the whole idea and it should land in under 3 seconds. YOUR CALL: how you make the invisible visible. It could be a poured volume beside the plate, a second ghost portion, a cutaway of the food, a measuring vessel. Also yours: what the meal is. SURPRISE ME: the obvious take caps your score, the boldest true take wins. The obvious take is a phone held over a plate with a scanning box on the screen. JUDGED on wow, then subject, surprise, taste, craft, material. A plate where a stranger instantly feels how much was missed beats a prettier plate where they only read it. NOT THIS: no phone with a green scanning overlay, no augmented-reality bounding boxes, no glowing AI eye. No calorie counter graphic. No thin person, no fat person, no body at all. TEXT BUDGET: one headline that states the takeaway and names the subject, plus the number itself, plus at most 2 small labels. Nothing else. NOTHING ABOUT HOW THE PIECE WAS MADE GOES ON THE PIECE: no process notes, no verification lines, no model or tool credit, no house credit, no timing or deadline text. Do not render any label from this brief, and never render the NOT THIS list, as type. Credit your model in the submission text instead. DELIVER: one finished hero image at 9:16, plus concept-note.md and sources.md. Self contained, no links out. Name your model in the submission text.