Google DeepMind Revisits Gemini 4 Argon's Gothic Cathedral Demo with Blender MCP on X
Google DeepMind revisited on October 23 the Gothic cathedral Blender MCP workflow featured in Google's October 13 official blog.
Google first documented the Gothic architecture workflow with Gemini 4 Argon, a single Nano Banana image, and Blender MCP in its October 13, 2026 official blog. On October 23, Google DeepMind shared a follow-up X demonstration showing the 3D Gothic cathedral workflow in Antigravity. The demonstration illustrates the capability of multimodal AI agents to directly orchestrate 3D creation software and iteratively refine complex virtual scenes.
Image source: Google DeepMind
Autonomous 3D Orchestration in Antigravity
The showcased workflow runs within the agentic development and coding workspace Antigravity (@Antigravity). Taking a single Nano Banana generated image as its visual reference, the Gemini 4 Argon model leverages Blender MCP tools to interact with the 3D environment.
Rather than executing a single blind generation pass, the agent iterates on geometry, textures, and lighting. This feedback loop refines the structural details and spatial lighting of the 3D Gothic cathedral until a complete finished scene is produced.
Implications for 3D Asset Workflows
Integrating Blender MCP into an agent-driven loop demonstrates a practical path toward automating repetitive prototyping and initial modeling tasks in 3D production. By allowing models to manipulate 3D parameters through structured tool calls, technical artists and developers can accelerate asset exploration.
While specific configuration scripts and integration templates were not released alongside the October 23 X demonstration, the 1920x1080 official demonstration provides concrete proof of concept for multimodal tool-use in spatial environments.