Name: Thom Vision
Author: Ic1558

Thom-Vision Skill

This skill equips Thom (Studio Agent) with "Eyes" to understand architectural visual data and "Hands" to guide generation.

Capabilities

Semantic Plan Reading (FastVLM)
- Input: Clean Plan Image (from S1).
- Action: Identify room zones, structure (columns/walls), and suggest circulation paths.
- Output: Semantic JSON/YAML description of the space.
Visual Mood Analysis (FastVLM)
- Input: Reference Image / Moodboard.
- Action: Extract material palette, lighting conditions, and architectural style.
- Output: scene_brief.yaml fragment.
Visual Control (ControlNet)
- Input: Line Art Plan + Prompt.
- Action: Constrain the diffusion generation to strictly follow the plan lines.

This skill equips Thom (Studio Agent) with "Eyes" to understand architectural visual data and "Hands" to guide generation.

Semantic Plan Reading (FastVLM)
- Input: Clean Plan Image (from S1).
- Action: Identify room zones, structure (columns/walls), and suggest circulation paths.
- Output: Semantic JSON/YAML description of the space.
Visual Mood Analysis (FastVLM)
- Input: Reference Image / Moodboard.
- Action: Extract material palette, lighting conditions, and architectural style.
- Output: scene_brief.yaml fragment.
Visual Control (ControlNet)
- Input: Line Art Plan + Prompt.
- Action: Constrain the diffusion generation to strictly follow the plan lines.