AI Scene Composer

The AI Composer is Yugma's primary interface — a chat panel where you describe what you want and the AI Director builds it.

How It Works

1. Type a description: "Build a sci-fi hangar with neon lighting"
2. The AI Director plans the scene (for complex requests)
3. It calls tools to create objects, set materials, adjust lighting
4. Objects appear in the viewport in real-time
5. You can iterate: "Make the walls darker" / "Add 6 pillars in a circle"

Modes

Mode	Temperature	Use case
Creative	0.7	Artistic scenes, varied output, mood pieces
Precise	0.2	Engineering layouts, exact dimensions, consistent

Toggle via the mode button in the AI panel header.

Scene Context

When "Scene On" is active, the AI sees your entire scene — object positions, materials, tags, and relationships. This enables:

• "Move that sphere to the left"
• "Make everything more metallic"
• "Add pillars matching the existing style"

Spatial Patterns

The AI handles arrangement requests with precomputed positions:

• "Arrange 8 cubes in a circle of radius 5"
• "Create a 3x4 grid of spheres"
• "Stack 6 cylinders"
• "Place 12 objects in a spiral"
• "Scatter 10 boxes randomly"

Positions are computed client-side (the AI doesn't do trigonometry) and injected into the prompt.

Style Memory

The AI analyzes your scene's existing materials and maintains consistency:

• Dominant color palette
• Roughness/metalness averages
• Style classification (industrial, neon, minimalist, etc.)

17 AI Tools

Tool	What it does
add_object	Create a new primitive or GLTF object
update_object	Change position, rotation, scale, material, name, tags
remove_object	Delete an object
set_environment	Change HDRI, fog, ambient lighting
clear_scene	Remove all objects
animate_object	Add A→B keyframe animation
duplicate_object	Clone N times with offset
align_objects	Align multiple objects on an axis
distribute_objects	Evenly space objects
focus_camera	Smoothly move camera to look at an object
apply_material_preset	Apply from 30 built-in PBR presets
search_select	Find and select objects by query
set_tags	Assign semantic tags
create_group	Group objects under an invisible parent
import_library_asset	Place a real model from the first-party libraries — Library v2 (world library; built in the browser at the requested quality, metre-accurate, floor-centred) first, the v1 industrial catalog (.ygm) when named or as fallback
search_library	Server-only discovery over Library v2: exact ids + metric dimensions for a query, family pack, country or industry
set_animation	Multi-keyframe animation with easing; jointTargets animates named joints on kinematic models
play_animation	Play an embedded animation clip on an imported GLB model

Cross-Session Memory

Your conversation history is automatically saved. When you return, the AI remembers your last session and can continue where you left off.

Rate Limits

All tiers share a burst cap of 30 AI requests per hour per user, on a rolling 1-hour window. This is an abuse guard, not your plan allowance.

Monthly allowances are per-tier: Free 100 AI requests/month, Premium 1,000/month, Ultimate 10,000/month.