How to spot AI art in games: Second Wind's checklist for cutscenes, sprites and 3D models

Second Wind's Design Delve sorts the tells of AI-generated game art by asset type. The checklist is useful, with one big limit: it only catches the careless cases.

Cole Walker

A magnifying lens over a glowing painting whose edges blur into static

Spotting generative AI in a game is less about one giveaway and more about a set of habits the tools leave behind. That's the argument of Second Wind's Design Delve episode on AI art in games, a 13-minute guide that sorts the signs by type of asset: cutscenes and video, 2D art, and 3D models. The host opens by explaining that the subject came up after he unintentionally played an AI-assisted indie game during a Steam Next Fest stream.

The structure is the useful part. Each kind of asset is made differently, so each fails in its own way.

Cutscenes and video

From 1:08, the video looks at AI-generated cutscenes. The signs it lists are short clips, subjects framed in the center of the shot, and blur that spreads into the edges of the frame.

The logic is straightforward. A video generator produces short clips, and it concentrates detail where the subject is, so the periphery is where it tends to give up. The host also argues around 1:17 that cost is the motive: AI cutscenes are cheaper than hiring a studio that specializes in them.

2D art

From 2:56, the video moves to 2D assets. Here the tells are a generic art style, lighting that comes from inconsistent directions, and a lack of hand-made sprite animation.

The last one is worth understanding. An image generator can produce a single convincing frame, but animation needs many frames that agree with one another. When a game's characters look detailed but barely move, that mismatch is the clue.

3D models

From 4:21, the episode covers 3D assets made through image-to-3D pipelines. It points to lighting baked into the textures, an artificial glossiness, and poor topology, the underlying mesh structure that artists normally build so a model deforms and shades cleanly.

Baked lighting is the most practical of these to check. If a shadow on a model stays put when the light in the scene moves, it was painted into the texture, not calculated by the game.

What the video wants done about it

The back half shifts from detection to policy. Around 8:16, the host says his main ethical objection is that the models are trained on artists' work without consent or payment. From 10:23, he applies survivorship bias to the question: the AI use people notice is the clumsy kind, which says little about how much goes unnoticed, especially in big-budget games.

That point matters for anyone using this checklist. These signs catch careless use. They can't prove a game is free of AI, and polished use won't show them.

His conclusion, from 11:37, is that detection shouldn't fall on players at all. He argues for mandatory disclosure, and for storefront filter tags, similar to the filters stores already offer for adult content, so players can exclude AI-made games if they want to.

How to use it

Treat the video's list as a set of flags, not a verdict. One sign on its own, such as centered framing or a generic style, describes plenty of human-made work too. Several signs together, in the same asset, are a stronger case. And for a purchase decision, a store's disclosure section is better evidence than any amount of squinting at screenshots, which is exactly why the video ends where it does.

Watch the video

Sources

More from Xarmo News