ChatGPT + Blender MCP Built This From Scratch

My wife's a game dev and asked me to help her put together a hospital corridor environment. My first thought was to use one of the image-to-3D generators, but then I wondered if an LLM could just drive Blender directly through MCP instead.

Quick disclaimer: I'm not a 3D artist. I can't model, UV unwrap, or light a scene manually. I'm an art director, so my skill set is prompting and directing, not hand-crafting geometry.

What you're seeing in the video is GPT-6 Astra using Blender MCP to build the entire environment from scratch: the corridor architecture, modular doors (open/closed/ajar variants), patient room interiors, hospital props (beds, IV stands, carts, wheelchairs, etc.), nurse station, lighting setup, and a short cinematic flythrough animation, all from text prompts, no external assets.

Happy to do a full write-up or step-by-step on setting up Blender MCP and the prompting approach if people are interested. Just let me know.

- Blockout stage: Build a realistic, explorable hospital corridor from scratch, roughly 35 meters long with a modular doorway system, alternating door states, a nurse station, and daylight windows at the far end, using an organized collection structure and a cinematic camera setup; construct the architectural blockout only and pause for approval before adding props or materials.

- Detailed dressing stage: Working within the approved corridor, add fully modeled reusable door panels and hospital props (beds, gurneys, wheelchairs, IV stands, carts, oxygen trolleys, and more), dress the four most visible patient rooms with partial interiors, upgrade the nurse station with desks and equipment, compose props deliberately across foreground, midground, and background, add environmental storytelling touches (loose papers, a displaced ceiling tile, mild wear), and refine the lighting to match the reference image before saving and rendering a preview still.

- Camera animation stage: Without altering any existing geometry or props, add a short 8-second forward camera push through the corridor using the existing camera's own orientation, keep the motion smooth with eased acceleration and only extremely subtle handheld-style sway, verify the path is clear of obstructions with test frames, then render and save the final animated video.

- Relighting stage: Keeping all geometry, props, and camera animation untouched, rework the lighting, materials, and color management to shift the scene from flat and overlit to a moodier, restrained cinematic look, with darker foreground pools, uneven ceiling fixture intensity, warm light spilling from a few patient rooms, a bright but non-blown-out far window, adjusted material roughness values, AgX color management, and very subtle volumetric haze, stopping after three lighting test frames are rendered for review.