Spline's Omma AI Generates Interactive Educational Prototype from Two Images.

A new community prototype demonstrates how Spline's Omma AI can turn simple reference images into interactive educational experiences. Joseph Azar, a designer working with the platform, built a functional elementary school lesson about ant habitats using Omma's multi-agent workflow and custom shader capabilities.

The project generated 3D models of an anthill's exterior and cross-section from two reference photos, then added sprite-based clouds and a custom shader mask to reveal interior chambers. According to Azar's post, AI acceleration allowed him to finalize the prototype quickly, though some educational details still need review.

Omma AI, launched by Spline in March, orchestrates code, image, and 3D generation agents simultaneously. For this anthill project, the system processed two input images: one showing the mound's exterior, another depicting the tunnel system in cross-section. The platform then generated matching 3D geometry for both views.

The shader mask implementation reveals a technical depth beyond basic 3D modeling. Custom shaders let developers control how surfaces render, and in this case, the mask selectively shows the anthill's interior chambers while keeping the exterior visible. This approach mirrors techniques used in architectural visualization and medical imaging, where revealing internal structures matters more than photorealistic surfaces.

Sprite-based clouds handle atmospheric effects without heavy rendering costs. Sprites are 2D images that always face the camera, a technique game developers use when polygon count matters. For web-based 3D experiences where performance determines whether users stay or bounce, this optimization makes sense.

Interactive 3D lessons solve a problem static textbooks can't: spatial understanding. Students grasp how ant colonies organize vertically when they can rotate the model, zoom into chambers, and see connections between tunnels. The format works for any subject where structure matters, from cell biology to urban planning.

Omma's natural language interface lowers the barrier for educators who lack 3D modeling skills. A teacher could describe a concept, reference a few images, and get a working prototype without opening Blender or learning WebGL. The system handles Three.js, React Three Fiber, and WebGPU under the hood, as previously demonstrated in path tracing experiments.

Azar noted the educational content needs review, pointing to a gap between technical execution and subject matter accuracy. AI can generate 3D geometry and code quickly, but it doesn't verify whether ant colonies actually organize their chambers the way the model shows. That validation step still requires human expertise.

Omma projects deploy to live URLs with custom domain support, using the same infrastructure that powers production web apps. The platform supports CSV, JSON, and DOC file imports alongside GLB and GLTF 3D models, letting creators connect real data to visual experiences.

The multi-agent architecture runs tasks in parallel rather than sequentially. When a user describes a scene, one agent writes code while another generates textures and a third builds geometry. This parallelization cuts iteration time from minutes to seconds, though it also means debugging requires understanding which agent produced which output.

Community remixing lets users fork existing projects as starting points. Someone could take Azar's anthill demo, swap the ant colony for a beehive, adjust the shader to show hexagonal cells instead of tunnels, and deploy a new lesson without starting from scratch. That workflow matches how developers use GitHub, but applied to 3D educational content.

Spline positions Omma as a tool for landing pages, apps, and games, but this prototype suggests a niche in educational technology. Schools need interactive content. Teachers lack time and technical skills to build it. AI-assisted 3D generation could fill that gap, assuming the accuracy problem gets solved.