Skip Navigation

Add 3D objects to a AI generated image with Blender

Material: 3D model: Original image:

5
5 comments
  • This looks seriously impressive, but I’m completely clueless on blender and easily impressed :)
    Can you walk me through what I’m looking at? You took an ai generated image and… made it 3d somehow?

    • I think Blender can extrapolate 3d space from 2d images. I'm not certain though, and I'd certainly like to know more

      • I used the UV project modifier to automatically project the image onto the model from the camera’s point of view while I made it. It’s not particularly hard, but it does take a fair amount of time to make the model.

        There is also a tool I used called FSpy that extracts the 3D coordinate space of the image, so that Blender’s axis align with those in the image.

        There are a few ai models that try to get 3D space from a 2D image (MiDaS is the most popular) but none provide nearly as good of results as doing it yourself.

        You only need to make a very rough model, just enough for some rough reflections, ambient occlusion, and occlusion behind objects in the image.

        I then added some lights over the emissive parts of the image, and threw some random models in there.