SciFiShooter
Autonomously Generated by AI — Unreal Engine 5.7
A single text prompt produced a complete third-person sci-fi shooter: production-ready C++ classes, a PCG alien terrain graph, Blueprint scripts, and a fully composed level — without a single line of manual code.
$ python3 main.py "Create a third-person sci-fi shooter with a player character that can dash, and a PCG-generated alien terrain."
Generated C++ Code
Production-ready Unreal Engine C++ with proper UCLASS/UPROPERTY/UFUNCTION macros, Enhanced Input integration, and collision-aware dash mechanics.
"token-keyword">void "token-class">ASciFiShooterCharacter::Dash()
{
"token-keyword">if (!bCanDash) "token-keyword">return;
"token-keyword">if (Controller == "token-keyword">nullptr) "token-keyword">return;
"token-keyword">class="token-comment">// Get forward direction from control rotation (yaw only)
"token-keyword">const "token-class">FRotator ControlRot = Controller->GetControlRotation();
"token-keyword">const "token-class">FRotator YawRot(0.f, ControlRot.Yaw, 0.f);
"token-keyword">const "token-class">FVector DashDirection = "token-class">FRotationMatrix(YawRot).GetUnitAxis("token-class">EAxis::X);
"token-class">FVector StartLocation = GetActorLocation();
"token-class">FVector TargetLocation = StartLocation + DashDirection * DashDistance;
"token-keyword">class="token-comment">// Collision sweep to prevent clipping through geometry
"token-class">FCollisionQueryParams QueryParams;
QueryParams.AddIgnoredActor("token-keyword">this);
"token-class">FHitResult HitResult;
"token-keyword">bool bHit = GetWorld()->SweepSingleByChannel(
HitResult, StartLocation, TargetLocation, "token-class">FQuat::Identity,
"token-class">ECC_Visibility,
"token-class">FCollisionShape::MakeCapsule(
"token-class">GetCapsuleComponent()->GetScaledCapsuleRadius(),
"token-class">GetCapsuleComponent()->GetScaledCapsuleHalfHeight()),
QueryParams
);
"token-keyword">if (bHit)
TargetLocation = HitResult.Location - DashDirection * 10.0f;
TeleportTo(TargetLocation, GetActorRotation(), "token-keyword">false, "token-keyword">true);
"token-keyword">class="token-comment">// Start cooldown timer
bCanDash = "token-keyword">false;
"token-class">FTimerHandle TimerHandle_ResetDash;
"token-class">GetWorldTimerManager().SetTimer(
TimerHandle_ResetDash, "token-keyword">this,
&"token-class">ASciFiShooterCharacter::ResetDash, DashCooldown, "token-keyword">false);
}Procedural Content Generation Graph
The agent designed a complete PCG graph with 11 nodes — from Perlin noise heightmap generation to slope-filtered vegetation scatter. Hover over nodes to trace the data flow.
PCG-generated alien terrain: Perlin noise heightmap, bioluminescent vegetation scatter, crystalline rock formations — all driven by the generated graph.
9-Phase Autonomous Pipeline
From a single prompt to 44 game-ready files — the agent orchestrates planning, code generation, Blueprint scripting, PCG design, and scene composition without manual input.
System Architecture
Seven Python modules form the agent's core, each responsible for a distinct layer of the game development workflow.
State machine that processes prompts, coordinates all generators, and manages the phase lifecycle.
OpenAI API wrapper with structured JSON output parsing and retry logic for reliable generation.
Produces UCLASS headers and .cpp implementations with correct UE5 macros and module dependencies.
Designs PCG graph specifications and generates Python Editor scripts to build them in UE5.
Creates Blueprint assets via UE5 Editor Python scripting, wiring variables and components.
Places actors, configures Lumen lighting, post-processing, NavMesh, and PCG volumes.
Generates .uproject, Build.cs, Target.cs, VS2022 workspace, and .editorconfig files.
Parses UnrealBuildTool logs, diagnoses MSVC errors with LLM, and applies targeted source patches.
Jinja2 templates for all UE5 C++ patterns: Characters, Controllers, GameModes, PCG nodes.
Generated Project Structure
44 files across 8 directories, ready to open in Unreal Engine 5.7.
More AI-Generated Games
The same autonomous agent — a different prompt each time. From dark fantasy dungeons to anti-gravity racing circuits, the agent adapts its C++ architecture, PCG graphs, and Blueprint scripts to fit any genre without manual configuration.
python3 main.py "Create a top-down dark fantasy RPG with a spell-casting knight, PCG dungeon generation, and skeleton enemy AI."
python3 main.py "Build a futuristic anti-gravity racing game with hover physics, boost mechanics, and a procedural cyberpunk city track."
python3 main.py "Generate a first-person survival horror game in an abandoned research facility with bioluminescent fungal enemies and AI stalking behavior."
python3 main.py "Create a 6-degrees-of-freedom space combat simulator with plasma cannons, target locking, and a procedural asteroid field."
All generated autonomously from text prompts — zero manual code written.