Slop Sense Launches Browser Game That Asks Players to Spot AI-Generated Images

Lab to AGI has released a browser-based game called Slop Sense that shows players a series of pictures and asks them to decide which came from a camera and which came from a generative model. The project lives at slop-sense.labtoagi.com and requires nothing beyond a web browser: no account, no payment, no software install. After each guess the site tallies how often the player got it right, turning image forensics into a few minutes of point-and-click.

The organization behind it describes the exercise as a way for people to calibrate their own judgment rather than as a formal evaluation of machine vision. That distinction matters. The site does not publish model weights, does not run a leaderboard, and does not claim to measure detector performance. It simply puts modern generators in front of ordinary users and lets them find out where their instincts hold up and where they fail.

The premise has grown harder to satisfy as image models have improved. Artifacts that once betrayed synthetic pictures, such as malformed fingers, garbled lettering and physically impossible reflections, appear far less often in current outputs. Detection research has not kept up. Academic evaluations routinely post strong numbers on the generators they were trained against, then lose much of that accuracy when confronted with an unfamiliar model, a weakness that makes any single benchmark number hard to trust.

Players who tried the game reported mixed results. Some said they scored comfortably above random guessing, while others conceded that portraits, wide landscapes and shots with heavy background blur left them unsure. A recurring point in the discussion was hindsight bias: once someone is told an image is fake, the giveaway details suddenly seem obvious, which means self-reported scores say less about real-world skill than they appear to.

For teams building with generative models, the exercise points at a shift already underway in how the industry handles authenticity. Watermarking, provenance standards such as C2PA and platform-level labels are the mechanisms regulators and standards bodies are pushing, in part because asking viewers to spot fakes does not scale. A casual game will not resolve that gap, but it does move the question from research papers to the people who scroll past generated images every day.

The obvious next step would be aggregate data. If Lab to AGI releases accuracy figures broken down by image category and player volume, it could give a rough public read on which visual subjects still fool most people and which no longer do. Until that happens, Slop Sense remains a personal test with no scoreboard beyond the browser tab.