When Crimson Desert launched last week, players did what players do: they started combing through the vast open-world RPG looking for details to appreciate or mock. What they found instead was something harder to ignore: in-game paintings that looked unmistakably AI-generated, featuring horses with too many legs, human faces that morphed into each other, and anatomical proportions that defied logic.
Three days later, developer Pearl Abyss issued an apology. Yes, the studio confirmed, some 2D visual props in the final release were created using generative AI tools during early development. According to Pearl Abyss, these assets were meant to be temporary; they were supposed to help the team "rapidly explore tone and atmosphere in the earlier phases of production" before being replaced with proper artwork. The plan, the studio said, was always to swap them out before players saw them.
Some just slipped through the cracks.
This explanation, while plausible on its face, triggered something unusual: a collective eye-roll from the rest of the gaming industry. Hundreds of game developers took to social media to demonstrate exactly what temporary placeholder art should look like. And it was not, they pointed out, something that requires an AI algorithm.
Josh Sawyer, studio design director at Obsidian Entertainment and a lead designer on Fallout: New Vegas, shared placeholder art from the studio's 2022 release Pentiment. One image featured Bambi hanging upside down. Another was simply the words "Guy Sux" in MS Paint. As Sawyer noted in his post, effective placeholder assets must look "obnoxiously temporary, so obvious that no one would mistake it for the final asset." Otherwise, he explained, "if you use a temp asset that seems passable, it may stay there."
That last sentence captures the heart of the problem. Placeholder assets in professional game development follow a simple principle: make them so ugly, so obviously temporary, that there is zero chance they survive to launch. Studios have used the Doge meme as icons, hot pink character models, their CEO's face, bright neon colours, stick figures, and crude doodles. The entire point is to make replacement non-negotiable.
Other developers piled in with examples that were equally crude and intentionally silly. Some studios used placeholder textures they found on their hard drives. Others generated them in minutes using the simplest available tools. The consensus was clear: you do not need an AI image generator to create something obviously temporary.
The deeper issue here extends beyond artistic choices. Pearl Abyss' statement suggested that the AI assets were created during early development and should have been replaced "following final work and review by our art and development teams." The studio also acknowledged it "should have clearly disclosed our use of AI" on Steam, where Valve has required such disclosures since early 2024.
This is not the first time a major publisher has made similar excuses. Late last year, Sandfall Interactive was stripped of its Game of the Year and Debut Game awards from the Indie Game Awards for the use of generative AI in Clair Obscur: Expedition 33 for placeholder textures that were mistakenly left in the game. The justification was identical: these were meant to be temporary.
For Pearl Abyss, the stakes are not insignificant. The game surpassed two million units sold within its first 24 hours, with the studio citing strong global demand across all platforms. A AAA release of that scale, with teams of hundreds, should have robust systems to prevent shipping with placeholder content of any kind, let alone obviously problematic AI images that contradict the game's stated art direction.
The studio has promised a comprehensive audit of all in-game assets and plans to replace affected content via patches. It has also committed to reviewing internal processes to prevent future oversights. These are reasonable commitments, but they also raise a question: if the studio's art and development teams were conducting final review before launch, how did paintings with centaur-horse hybrids and anatomically impossible faces pass through?
The developer backlash suggests a broader scepticism. In an industry where budgets have ballooned and teams have grown, there is growing concern that quality control has become a casualty. The assumption that AI tools can serve as placeholder content, paired with inadequate replacement discipline, reflects not innovation but rather corner-cutting dressed up in technological language. It is easier to blame accidental oversights than to acknowledge that somewhere in the approval chain, the obvious was missed.
This matters not because placeholder assets in early development are inherently wrong, but because shipping them suggests failures at multiple levels: discipline, oversight, and basic quality assurance. Pearl Abyss may have had good intentions with its AI experimentation. But good intentions and a 2 million-copy launch do not excuse shipping a finished product that looks unfinished.