SCENARIOS
The drift meet you fly through is not AI. It is somebody's actual run.
Five live scenarios, scattered across Los Santos and Blaine County — a drift meet trading corners under coloured neon, jet skis running the coast, stunt planes making passes over the airport. Every one of them is a recording of a run a real person flew or drove, replayed on a clock that every player on the server shares.
HOW IT WORKS
Somebody drove it. We kept the drive.
A scenario is recorded rather than programmed. Somebody flies the plane or drives the car, the run is captured, and it is played back for everyone on a shared clock — as local entities on each machine, so nobody is simulating anybody else’s traffic. Several takes can be grouped into one scene and timed against each other, which is how a drift meet ends up being a meet rather than a set of cars near each other.
It sounds like a smaller idea than AI. It is a much better one, and the consequences are all things you notice while flying.
It moves like a person, because a person moved it
AI drifts like a machine solving a corner. A recording drifts like somebody drifting — the small correction, the slightly late entry, the recovery. That texture is the whole reason the scenes read as alive, and it is not something a behaviour tree produces.
Choreography, not coincidence
Several recordings can be arranged against one shared timeline, so two cars trade the same corner on every single lap rather than merely being in the same place. It is authored the way a shot is authored, which means you can learn a scene and plan a line through it.
They are solid
Not scenery. Misjudge a gap between two drift cars and you wreck, exactly as if you had hit a wall — which is what makes flying through one worth anything at all.
Boats and jet skis are real craft
Watercraft follow their recorded line as genuine physics objects rather than replayed positions, so wake, buoyancy and impacts are live. That is a per-scene decision, made where the physics is the point.
It costs almost nothing
This is the engineering reason the approach was chosen. AI needs pathfinding and a task scheduler, decides something every frame, and — the part that ends servers — one unlucky client has to simulate all of it for everybody. Playing back a recording locally does none of that, so scenes can be dense without costing you frames.
Every one is marked on the map
With an icon matching what is actually there, so finding something happening does not depend on flying over it by luck.
THE ROAD NOT TAKEN
We built the AI version first
An earlier version of this world populated itself with AI — crews, meets, events that woke up as you approached. It was retired, and the reason is structural rather than a change of taste: in a shared world one client ends up simulating the AI for everybody in range, and that client pays for all of it. It behaved like a robot and it cost frames.
Recording real runs solved both problems at once, and improved the thing we were actually chasing. The scenes look human because they are.
Go and fly through one
Scenarios are marked on the map, they run on their own schedule, and there is nothing to start or unlock — they are simply happening. Bring a camera: the cinematic airframe seats a second pilot on the gimbal.
CONNECT PLAY.GTAFPV.COM