Skip to content

ai-level-design-tools.md

Latest commit

 

History

History
126 lines (101 loc) · 5.32 KB

ai-level-design-tools.md

File metadata and controls

126 lines (101 loc) · 5.32 KB
Raw

AI-Powered Level Design Tools (OpenAI API Integration)

Purpose

This document outlines a practical tool suite for integrating AI into VibeRadiant map production using the OpenAI API. The focus is production speed, consistent quality, and safe editor-side automation.

Goals

  • Cut blockout, encounter, and polish iteration time.
  • Reduce build-break and logic-break regressions.
  • Keep AI changes reviewable, reversible, and deterministic.
  • Fit naturally into existing VibeRadiant command, selection, and undo workflows.

Proposed Tool Suite

1) Prompt-to-Blockout Generator

  • Converts a design brief into a playable blockout plan.
  • Produces room and route graph proposals, then brush or patch actions as a preview diff.
  • Applies only after user confirmation.

2) Encounter Composer

  • Populates selected areas with enemy, item, cover, and objective placements.
  • Optimizes for pacing constraints such as pressure, travel timing, and sightline balance.
  • Emits placement rationale and quick variant options.

3) Entity Wiring Assistant

  • Builds and validates trigger logic chains (target, targetname, relay patterns, objectives).
  • Auto-fixes missing keyvalues, broken references, and naming conflicts.
  • Integrates with issue-style diagnostics for one-click repair.

4) Texture and Shader Pass Assistant

  • Applies semantic material suggestions to selected surfaces or zones.
  • Normalizes scale, rotation, and alignment to style rules.
  • Flags shader misuse and proposes safe replacements.

5) Lighting Pass Copilot

  • Proposes light entities and keyvalues from gameplay intent (readability, mood, path guidance).
  • Iterates from compile logs and viewport feedback.
  • Supports fast pass and quality pass profiles.

6) Compile and Error Triage Agent

  • Parses compile output and maps errors to likely geometry or entity causes.
  • Suggests minimal, ranked fix actions.
  • Can batch non-destructive fixes in a single undo transaction.

7) Playtest Analytics Coach

  • Ingests telemetry (heatmaps, death locations, route frequency, objective failure points).
  • Ranks map edits by expected impact.
  • Generates targeted follow-up tasks for the next iteration.

Integration Architecture

Editor-Side Components

  • AI Tools panel for prompts, previews, and apply flow.
  • Command entries for each tool action.
  • Selection-aware context capture (active selection, layer, map mode, gamepack metadata).
  • Undo-safe operation batching for all applied edits.

AI Service Layer

  • Use OpenAI Responses API as the orchestration core.
  • Use function calling with strict structured outputs for edit operations.
  • Disable parallel mutating calls during map edits to avoid conflicting changes.
  • Use vector store retrieval for local design conventions, mapping standards, and gamepack rules.
  • Use background mode plus webhooks for long-running map-wide analysis.
  • Use batch processing for nightly audits or style conformance scoring.

Suggested Tool Functions (Examples)

  • create_brush_prism
  • create_patch_grid
  • move_selection
  • set_entity_keyvalue
  • connect_entity_targets
  • apply_shader_to_faces
  • run_validation_checks
  • propose_fix_batch

Each function should return typed, machine-validated payloads so every operation is deterministic and replayable.

Trust and Safety Guardrails

  • Preview-first workflow: AI proposes, user confirms.
  • Scope boundaries: selection, layer, or named zone only.
  • Hard limits: operation count, spatial bounds, entity class allowlists.
  • Auto-rollback: every apply is wrapped in one undoable command.
  • Audit trail: store prompt, context hash, tool calls, and applied diff summary.
  • Protected modes: disallow destructive global actions unless explicitly enabled.

Rollout Plan

Phase 1 (Highest ROI)

  • Compile and Error Triage Agent.
  • Entity Wiring Assistant.
  • Shared AI command pipeline with strict function outputs and preview UI.

Phase 2

  • Prompt-to-Blockout Generator.
  • Encounter Composer.
  • Selection and zone-aware apply constraints.

Phase 3

  • Texture and Shader Pass Assistant.
  • Lighting Pass Copilot.
  • Rule packs per game type and style profile.

Phase 4

  • Playtest Analytics Coach.
  • Continuous optimization loop from telemetry to prioritized map edits.

Success Metrics

  • Time-to-first-playable blockout.
  • Compile error count per iteration.
  • Manual fix count after AI-assisted apply.
  • Encounter balance iteration count to target metrics.
  • User acceptance rate of AI proposals.
  • Revert rate of AI-applied changes.

Cost and Performance Notes

  • Use smaller, faster models for interactive proposal drafts.
  • Reserve larger reasoning models for map-wide planning and triage.
  • Cache static context (style guides, schemas, conventions) to reduce token usage.
  • Run heavy analysis asynchronously and notify on completion.

Implementation Notes for VibeRadiant

  • Primary integration points are command registration, plugin dispatch, selection APIs, and undo transaction boundaries.
  • Keep initial implementation as a non-invasive plugin plus service bridge.
  • Promote stable operations into core editor commands only after reliability targets are met.

Summary

This tool suite is designed to accelerate production without sacrificing editor control. The core principle is constrained automation: AI handles high-volume planning and repetitive setup, while designers remain in control of final intent and quality.