Era3D: High-Resolution Multiview Diffusion Using Efficient Row-Wise Attention

The authors have set up a live demo on their HuggingFace page: https://huggingface.co/spaces/pengHTYX/Era3D_MV_demo

It's very easy to upload your image, click the button called "Generate Normals and Colors" and get multiple views of the generated 3d model populated below.

For full experience and mesh export, their Github repo is the way to go: https://github.com/pengHTYX/Era3D?tab=readme-ov-file

Update: here is an example of a generated Nokia phone that can hold its battery for at least a year: https://imgur.com/PzUfbW7

Don‘t miss a “Lewd statue of an angel texting on a cell phone” in text to 3D examples

The “image to 3d” examples seem to do well if the source image is an oblique view but not if it’s a direct straight-on front view.

I like that it couldn't decide if it was a ghost eating a hamburger or a hamburger that eats ghosts from that ambiguous prompt.

What's the SOA in photogrammetry for the amateur with modest devices?

Seems like this would be very useful for game dev. This looks amazing!

Dat is some dark magic