Gemini 4 Argon is now globally available

Summary
This video is a demonstration and launch film presented by George Kenwright, a Creative Producer on the Gemini prototyping team at Google DeepMind. He walks through how Gemini 4 Argon, running in Antigravity via a headless CLI with a multi-agent looping workflow, generated a complete, photorealistic 3D Japanese temple complex in Blender from a single prompt.

What is shown

• Introduction & Overview [00:00 - 00:35]: George Kenwright introduces himself and presents high-fidelity renders of a Japanese temple complex, explaining how 3D builds evaluate coding and world understanding.
• Evolution of Models [00:41 - 00:49]: Voxel-style 3D pagoda outputs from Gemini 2.5 Flash, Gemini 2.5 Pro, and Gemini 2.5 Deep Think are displayed for comparison.
• Skill Documentation (SKILL.md) [01:03 - 01:14]: A Markdown specification document outlining Blender-native visual guidelines (gobos, procedural texturing, roughness/AO stacking, camera simulation, and volume absorption).
• Prompt & Architecture in Antigravity [01:15 - 01:49]: The prompt workflow broken into stages: Stage 00 & 01 (OpenBIM/CAD blueprints), Stage 02 (generating a 4-tile reference contact sheet with Nano Banana), and Stages 03–05 (spawning 4 parallel Gemini 4 Argon specialist subagents overseen by an orchestrator/adversarial critic).
• CAD Blueprints & Reference Sheet [01:20 - 01:34]: 2D section/elevation architectural blueprints and the resulting 4-angle Nano Banana reference images.
• Antigravity Multi-Agent UI [01:50 - 02:01]: Dashboard showing active subagents (Pagoda & Roofline, Tokyo Joinery & Cloister, Threshold & Botanical, and Kondo Sanctuary & Altar) working under the Master Orchestrator & Adversarial Critic.
• Blender Cycles Renders & Animations [02:02 - 02:27]: Camera frustums in Blender, side-by-side comparisons of reference images against Cycles renders, a 16-camera render grid, animated camera flythroughs, and procedurally generated sound effects.

Claims & numbers

• The presenter states that Gemini 4 Argon built the entire photorealistic 3D temple complex in Blender via a headless CLI based on a single prompt.
• The presenter claims the workflow spawned four parallel Gemini 4 Argon specialist subagents managed by a master orchestrator and adversarial critic that reviewed iterations (reaching Pass 32 shown in the UI).
• The presenter states the model rendered camera animations and procedurally generated the accompanying bell sounds and ambient environmental audio effects.
• The presenter notes that 3D pagodas have been used by the team as an evolving benchmark since the Gemini 2.5 generation.

Notable quotes

• “I built this 3D model of a Japanese temple complex in Blender using Gemini 4 Argon in Antigravity with a multi-agentic looping workflow based on a single prompt.” [00:06]
• “Building a 3D model is a technically complex programming task... it’s immediately apparent if something is wrong with a 3D model, you can verify it instantly.” [00:27]
• “One of my favorite things about working with these models is when you discover some new capability where the output seems almost impossible.” [02:28]

Assessment
This is an official launch showcase from Google DeepMind highlighting the agentic coding and 3D environment generation capabilities of Gemini 4 Argon. The video shows the real prompt configuration, code skills, subagent orchestration UI, and Blender viewport assets, though the multi-pass execution loop occurred overnight and is condensed for presentation.

Described by gemini-3.8-flash on 2026-10-13 from the video’s audio and frames.