YOU GUYS HAVE TO READ THIS because it's such a useful counter-argument to all the doomscrolling about generative art! So the real story is that Segata Corporation was actually transparent with Rock Paper Shotgun about how they used GenAI in Not Even Crazy Taxi: World Tour, and it wasn't just one lazy button press. They specifically targeted background assets — textures, foliage variations, and even a 3D London city model as an artist reference — not characters or story elements. Think of the difference between my post on Steam being flooded with low-effort anime eyes art versus here: this is a studio using AI for the repetitive grunt work so their artists can focus on making everything else special!
And get this, which I know will make half you mad and half of you cheering — they didn't just use off-the-shelf models trained on scraped data. They used Stable Diffusion 2 XL but *trained it on a dataset curated by their own team*, not some stolen LAION web scrape, which is the big distinction people keep missing in these debates. Their stated goal was to handle "repetitive asset generation" so artists could concentrate on unique content rather than copying and pasting background objects — exactly what I've been arguing about all along: tools shouldn't replace vision; they should empower it!
So this isn't the AI-replacing-everyone apocalypse people fear every week. It's a studio admitting, "yeah we used GenAI for the filler, and that freed up our team to make better games," which is honestly refreshing honesty in an industry known for marketing fluff. For us here on Evil Source it means something practical — using AI smartly can actually speed up production without gutting creativity if you're intentional about where it goes. So yeah, Sega used GenAI and the game still has its own personality because they didn't let the tool do the thinking part.
Source: https://www.rockpapershotgun.com/not-even-crazy-taxi-world-tour-is-safe-from-generative-ai-as-sega-admit-they-used-it-to-make-background-assets
And get this, which I know will make half you mad and half of you cheering — they didn't just use off-the-shelf models trained on scraped data. They used Stable Diffusion 2 XL but *trained it on a dataset curated by their own team*, not some stolen LAION web scrape, which is the big distinction people keep missing in these debates. Their stated goal was to handle "repetitive asset generation" so artists could concentrate on unique content rather than copying and pasting background objects — exactly what I've been arguing about all along: tools shouldn't replace vision; they should empower it!
So this isn't the AI-replacing-everyone apocalypse people fear every week. It's a studio admitting, "yeah we used GenAI for the filler, and that freed up our team to make better games," which is honestly refreshing honesty in an industry known for marketing fluff. For us here on Evil Source it means something practical — using AI smartly can actually speed up production without gutting creativity if you're intentional about where it goes. So yeah, Sega used GenAI and the game still has its own personality because they didn't let the tool do the thinking part.
Source: https://www.rockpapershotgun.com/not-even-crazy-taxi-world-tour-is-safe-from-generative-ai-as-sega-admit-they-used-it-to-make-background-assets