Spline’s Spell Summons 3D Scenes from a Single Snapshot

Spline is no stranger to radiance field representations. In fact, it was among the first platforms to adopt Gaussian Splatting shortly after its initial release. Now, the team has significantly expanded its capabilities with the introduction of Spell, a tool that can generate a 3D scene from a single image. According to Spline’s own blog, Spell supports multiple radiance field methods, including both Gaussian Splatting and NeRFs.

Why emphasize this flexibility? While the “floor” of Radiance Fields is already quite high, these existing methods have much higher ceilings and are likely to become even more useful. It’s also overwhelmingly like that novel Radiance Field methods will continue to emerge— potentially sooner than many anticipate.

The output strongly resembles Google’s CAT3D and under the hood, Spell relies on diffusion modeling, a type of AI known for high-fidelity imagery. It emphasizes realism, multi-view consistency, and adaptability across diverse subjects—from people and everyday objects to environments and stylized 3D characters. While the company notes that Spell can extrapolate beyond its original training data, it also acknowledges there’s room for improvement in quality and consistency.

To develop the model, Spline’s team captured real-world images from various countries, creating a ground-truth dataset that helps Spell learn how objects and scenes appear from different angles. In parallel, they generated synthetic data using a range of 3D modeling techniques, some of which were custom-built. They also used licensed 3D assets from trusted third-party marketplaces—but none of Spline’s user data was employed in training.

Spell offers three main output options:

1. A simple animated clip panning around the generated scene.

2. A series of still frames taken from multiple angles.

3. Direct export as a Gaussian Splatting radiance field.

Described by Spline as a first step toward AI-driven, interactive graphics, Spell marries automated 3D generation with user-friendly editing tools. This could allow designers to build large-scale virtual worlds more quickly and flexibly. The company plans further refinements over the coming months, including new training checkpoints to enhance consistency, detail, and workflow integration.

Currently, Spell is available for a $99 monthly subscription, though free users can still view others’ creations. As the technology evolves, it promises to open up new frontiers in 3D creation and design, making advanced graphics more accessible than ever before.