General purpose frontier AI models are not yet proficient at turning 2D images into 3D models. Specialized AI models can perform this task with stunning accuracy and detail for many works, but frontier models are not yet strong at this task. Nevertheless, let's see what Gemini 3.8 Flash Code Execution can do to turn the engraving below from Jode's 1585 Thesaurus (the first picture Bible) into a 3D model. The resulting model is a crude Lego-like reconstruction, but interestingly, Gemini has added physics simulation to the model to make it dynamic, demonstrating what will increasingly become possible as frontier models are able to interact more closely with specialized AI models like 2D->3D reconstruction engines.
The source image:
A snapshot of Gemini's reconstruction:
