Capcom has revealed plans to evolve the RE Engine with AI features. That is roughly the extent of what has been said, and the announcement is worth noting precisely because the phrase is so wide.
The RE Engine is not an incidental piece of technology. It is the engine behind Resident Evil 7, the RE2, RE3 and RE4 remakes, Devil May Cry 5, Monster Hunter Rise and World, and Street Fighter 6 — which is to say, the engine behind a decade of work that is widely held up as the benchmark for how to remake a game well.
”AI features” is not one thing
The announcement will be read by a lot of people as “Capcom will generate art with AI.” It might mean that. It much more likely means several things, and they occupy very different positions on the question of whose job is affected.
Roughly in order from uncontroversial to contested:
Technical automation. LOD generation, UV unwrapping, collision mesh creation, texture compression decisions, lightmap baking optimisation, automatic retopology. Nobody enjoys doing these and nobody’s artistic voice is in them. This is where most engine-level ML already is, and where the productivity win is largest and least arguable.
QA and testing. Agents that play the game to find collision holes, soft locks, and places the player can escape the level. Capcom ships large open-ish worlds and this is enormously expensive to do by hand.
Animation tooling. Retargeting between skeletons, cleaning motion capture, in-betweening, foot-slide correction, generating transition animations between states. Genuinely hard craft work, partly automatable, and the automation is mostly welcome — the alternative is an animator hand-fixing a thousand transitions.
Upscaling and frame generation at runtime. DLSS-class technology, which is now table stakes and which the engine has to support regardless of anyone’s opinion about it.
Asset generation. Textures, materials, props, variations. This is where the argument is, and where “AI features in the engine” becomes a labour question rather than a tooling one.
NPC behaviour and dialogue. Generative characters. Interesting, mostly unsolved in a way that survives shipping, and the thing most likely to be in a tech demo and not in a game.
Until Capcom says which of these they mean, the announcement is not assessable. Reading it as the most alarming option is as unfounded as reading it as the most benign.
What makes RE Engine an interesting place to do this
Two things about the engine specifically are worth knowing, because they shape what AI features could usefully do.
It is an in-house engine with a tight art pipeline. Capcom’s remakes are praised for a very specific achievement: photogrammetry-heavy environment art that still reads as authored. They scan enormous amounts of real material and then art-direct it hard. That workflow — capture, clean, stylise, integrate — is full of exactly the laborious technical steps that automation suits, and it is also a workflow where the authored layer is the value.
And it is a performance-focused engine. RE Engine games are notable for running well on modest hardware, and Capcom has been willing to trade visual maximalism for frame rate. An engine with that culture is more likely to add AI for optimisation and authoring than for runtime spectacle.
Which is why we covered the Marvel Rivals art team’s hybrid NPR/PBR and VFX-based destruction last week as a related case: the interesting engine decisions in this generation are overwhelmingly about moving cost from runtime to build time, and that is the category most ML tooling actually lands in.
How to read announcements like this
A suggestion, offered because there will be a lot more of them:
Ask which step of the pipeline it replaces. “AI in the engine” is meaningless. “Automatic retopology in the engine” is a tool. “Texture generation in the engine” is a staffing decision.
Ask whether the output is editable. A tool that generates a mesh an artist can then adjust is an accelerator. A tool that generates a mesh nobody can adjust is a dependency.
Ask who it is aimed at. Engine features aimed at technical artists and at tools programmers have a very different character from features aimed at reducing headcount, and companies are usually reasonably clear about this when asked directly.
And watch what ships, not what is announced. Engine roadmaps are long, and the gap between “plans to evolve” and a feature in a shipped title is typically two or three years.