Logo image
Back
AI-assisted procedural environment generation in Unreal Engine 5: a comparative study with manual production workflows
Thesis   Open access

AI-assisted procedural environment generation in Unreal Engine 5: a comparative study with manual production workflows

Aayush Sanjay Singh
Master of Science (M.S.), Drexel University
Jun 2026
DOI:
https://doi.org/10.17918/00011411
pdf
Singh_Aayush_202647.31 MBDownloadView
PDF Open Access Open Access (License Unspecified)

Abstract

AI-assisted environment generation Blueprint visual scripting Large language models Procedural content generation Spline-based generation Unreal Engine 5
The creation of three-dimensional outdoor environments for games and digital media is one of the more time-intensive aspects of real-time production. As environments grow in complexity and visual fidelity demands rise, the cost of assembling scenes by hand has become a practical constraint, particularly during early-stage design when iteration speed matters most. This thesis investigates whether an AI-assisted procedural generation framework built inside Unreal Engine 5 can reduce that cost meaningfully, and what the practical trade-offs of such an integration look like. The implemented prototype takes a plain-language scene description as input, sends it to a large language model through a Python script, and uses the structured JSON response to configure procedural generation directly within the engine. A Blueprint Editor Utility Widget parses the model's output and passes the extracted parameter values to a dedicated controller actor, which coordinates terrain streaming, foliage and rock distribution through the Unreal Engine PCG framework, lighting preset application, and path or river construction along a runtime spline. Roads and dirt paths are rendered as decal projections onto the terrain surface; rivers are generated by spawning and configuring a Water Body River actor at runtime. All generation draws from a pre-prepared project asset library and operates within the engine's native toolset. The prototype was evaluated against two reference approaches: manual scene creation using Unreal Engine's foliage tools, and procedural generation configured by hand through the PCG framework without AI assistance. Across both a forest and a desert test case, the AI-assisted approach produced initial scenes far more quickly than either reference, with total time remaining substantially lower even when prompt refinement and manual adjustment were included. Asset modification through prompt update was also considerably faster than repainting or graph reconfiguration. One consistent technical observation emerged during river generation testing: a synchronization gap between the Water Body River's asynchronous landscape deformation and the PCG framework's heightfield sampling caused procedural assets to appear at the original terrain height rather than on the deformed surface. The behavior is tied to the Landmass brush system's asynchronous commit timing and is reported here as a practical finding about runtime procedural integration in Unreal Engine 5. The thesis frames the contribution as a proof-of-concept implementation study rather than a production-ready tool or a machine learning research contribution. The results suggest that a prompt-driven AI-assisted approach can meaningfully support rapid environment prototyping by reducing the configuration overhead that precedes artistic work, while manual authoring and parametric procedural tuning remain important for final scene refinement. The integration challenges encountered during development point to useful directions for future work in engine-native AI-assisted content generation.

Metrics

1 Record Views

Details

Logo image