Text-to-3D Model AI Generation

OpenAI recently announced Shap-E, a text-to-3D model tool. Although not the first (other prior text-to-3D modeling tools includes NVIDIA's Magic3D) it is in the AI news and I am sure that the question everyone is asking is, 'How will Holovision use this tool in his next Gotham City caper?' 'How good is the quality of the 3D objects Shap-E generates?'"

I started with the "Utah teapot" prompt which is one of the traditional models for testing 3D graphics and Shap-E added some deformities and unnecessary details to the surface of the Utah teapot. For some simple and common 3D objects Shap-E does a pretty good job. For example, this red toy truck below. If you use huggingface.co you can generate virtual 3D objects for free with no software download using text prompts in your browser. You can designate the color of the object in the prompt and once generated in the browser you can rotate the object in any direction and zoom in and away from the 3D object.

If you want to edit the generated 3D object in 3D graphics software like Blender you can download the 3D object as a GLB file by clicking on the download icon in the upper right hand corner of the preview window on huggingface.co.

This is supposed to be a stop sign. Similar to the 2D images generated by AI image generators like Stable Diffusion the AI has difficulty adding readable text to the 3D objects it generates.

The dog looks a little like a hamster but it is a dog.

Shap-E can generate a chessboard with all the squares. This must have been from one of those Cold War games between an American and a citizen of the Soviet Union because it looks like the chess pieces were nuked.

Shap-E can generate a chessboard with pieces but when I used the "alien solar system" prompt Shap-E just generated a ringed planet.

This is supposed to be "a beautiful princess" but rather than being from a fairy tale her face seems to be from a creepypasta.

Another attempt at generating a 3D human character. 2D image generating AIs seem to handle 2D faces well but not 3D faces. I am optimistic this is only temporary and that the software will improve.

I know that the original appearance of Batman doesn't fall into the public domain until 2035 (under current U.S. law) but the prompt I used on Shap-E to make Batman look sad and pathetic is just too good to not share. They can stop me from writing about my Gotham City capers but they can't stop artificial intelligence! Just look at this 3D model of "loser Batman" and laugh maniacally with me.