XNews

GPT-6 Astra Shows Diverse 3D and Creative Generation Demos

GPT-6 Astra Shows Diverse 3D and Creative Generation Demos

Multiple demonstrations of OpenAI's GPT-6 Astra (via Codex) show its ability to generate complex 3D scenes, interactive games, and more. One example: turning a single sentence into an editable 3D house in Blender and then exporting to Unreal Engine 5 with interactive features (doors, light switches, coffee machine). Other demos include a V8 engine 4-stroke cycle visualization, human anatomy with 2,234 interactive 3D models, a robot painting task learned through trial and error, and a personal physiotherapy system linking wearables with biomechanical models. A 2D game was also built using GPT-image-2.5 sprites. The model shows self-checking and iterative refinement.

Key insights

Astra Can Build Complex 3D Scenes from Simple Prompts

A developer used a single sentence to have Astra build a detailed 'Solace' house in Blender: structure, furniture, plants, lighting, and camera. Astra iterated with self-checks, fixing issues like plant clipping. The scene was then exported to Unreal Engine 5 with first-person controls, interactive objects, and volumetric lighting. The final Blender render was 35 seconds at 1080p with 900 frames.

Diverse Applications: Education, Robotics, Health

Another demo showed a V8 engine cutaway with interactive 4-stroke cycle animation showing crank, camshaft, ignition timing. A human anatomy model with 2,234 interactive 3D models covering bones, muscles, organs, blood vessels, and nerves was built. A robot arm learned to paint the Golden Gate Bridge using camera feedback, improving over attempts. A personal physiotherapy setup connected wearable devices to a biomechanical model to track muscle pain sources and adjust exercises.

Astra Also Excels at 2D Game Creation with Image 2.5

A user combined GPT-6 Astra and GPT-image-2.5 to create a 2D platformer game: sprites for characters and enemies generated as sprite sheets, background, platforms, and sound effects were all produced. This shows Astra's ability to coordinate multiple generation tools.

Self-Correction and Iteration Are Key Features

In multiple demos, Astra actively checked its own work: finding that the Saturn ring was clipped, the ship hull too plain, or planets competing for attention, and then fixed them. The Blender workflow involved rendering previews, checking results, and adjusting.

Source posts