Introducing Spell: Spline's exciting new tool for creating realistic 3D scenes from images

Popular 3D platform Spline has unveiled Spell, a new model designed to generate 3D worlds directly from images. This browser-based tool can create entire 3D scenes, referred to as “Worlds”, within minutes, maintaining consistency with the original image input. The generated scenes are represented as volumes that can be rendered using Gaussian Splatting or alternative methods like Neural Radiance Fields (NeRFs).

Spell operates as a diffusion model, capable of producing 3D worlds with realistic multi-view consistency across diverse categories, including people, objects, environments, and 3D characters. It can render images from various angles with high precision and detail, generate controlled camera paths, and simulate physical material properties such as reflections, refractions, and surface roughness. Additionally, Spell can simulate camera properties like depth of field and manage camera/object intersections for realistic internal surface views.

The model emphasizes physical consistency, simulating camera intersections rather than relying on interpolation, to maintain a realistic visual flow. Currently, Spline is offering Spell in an early access phase with limited availability and a high price point, targeting early adopters to gather user interaction insights while managing GPU costs.