Large-Scale Levels Workflow
When working on AAA or open-world levels containing tens of thousands of GameObjects, sending the entire scene hierarchy to remote AI models will exceed API token limits and increase network latency. Follow these rules to optimize performance.
1. Selection-Based AI Scope
Instead of requesting a reorganization of the entire scene, isolate specific sectors:
- In the AI Director tab, set the Scope dropdown to
Selected Objects. - In your Hierarchy window, select only the group of items you wish to sort (e.g., select the 50 meshes representing a specific outpost).
- Type your prompt:
"Create subfolders for barricades, crates, and lights, and organize these items inside." - Execute the prompt. The LLM only processes the metadata of those selected objects, completing the task in under a second for a fraction of the cost.
2. Setting Max Scan Depth
For complex nested rigs or characters, prevent the scanner from diving too deep:
- Go to the Settings tab of the Dashboard.
- Reduce the Max Scan Depth slider to
3or4(down from default8). - This forces the scanner to ignore deep bone chains, renderer materials, and nested UI buttons, scanning only major parent layers.
3. Disabling Coordinate Matrices
If you only need structural/naming categorization rather than spatial grouping (like “group items based on how close they are”):
- In Project Settings, toggle Exclude Transform Matrices to
ON. - This stops the tool from generating
x, y, zcoordinates for every GameObject in the payload. - Decreases your prompt token count by up to 50%, avoiding API limit errors and cutting costs in half.
4. Segmented Atmospheric Control
Do not run the Atmospheric Controller when multiple lighting groups are present (e.g. Day/Night variations).
- Rename the directional lights clearly (e.g.
Light_DayandLight_Night). - Deactivate the alternative light set before running the mood prompt to ensure the AI modifies the correct active directional light.