The other day I used the following workflow to use AI to generate a 2D person in a specific perspective. This can be an option until there is an AI that creates detailed 3D persons. The process is a bit convoluted with many steps. If there is a shortcut, please let me know.
DALL·E: Created a man in a Gho, the Bhutanese formal dress. I tried to tell DALL·E to give me a perspective slightly from above, but to no avail. As we all know, these AIs do what they want, and our prompts are more or less suggestions.
Magnific.ai: Upscaled the image four times to make it look photorealistic. This step is not necessary. I did it because originally I thought I can use the person even if the perspective is off. But it looked weird in the scene.
Rhino3D: Positioned the model in a perspective that matches that in D5. Took a screenshot. It would be better to do that in D5 itself, but I didn’t figure out how to import a textured model there, plus D5 does not support GLB out of the box.
Magnific.ai + Clipdrop + ImageMagick: This is the magic part. Magnific.ai makes the blurry, unusable screenshot from Rhino look photorealistic. Note that it can mess up hands, but there are no hands here. The dress could look a bit cleaner, and maybe that is possible with good prompt engineering. Finally, I used Clipdrop and ImageMagick, as above, to create a cropped image with transparent background.
Honestly, really easily. You just upload the D5 image to magnific.ai,you don’t even have to prompt anything as it’s able to understand what’s in the picture, and after a minute or so, it has upscaled your image and added a lot of detail to everything ! I hope something like that could be integrated natively in D5.
See here how detailed the bags are after upscaling
When I tried piping my entire rendering through magnific.ai, a few weeks back, I was greeted with artifacts such as tiny people sitting in a tree, like ferries. Granted, the rendering was of a very unconventional building that integrates a living tree. In the end, I couldn’t use it, even after playing with the sliders and reducing its imagination (forgot what they are called).
Upon second look, there are plenty of artifacts in the hand bag example too, for example a highlight that was turned into a crumbled notebook (?), a headphone earcup that looks odd, a frown instead of a smile, a bent yellow stick instead of pens. Also, it did something weird to the mesh behind the shelf.
I suppose you dont want to have Maginific AI or other AI’s out there to resolve everything for you, Is good to add a bit more realistic looking features to your final image but you should not relly only on this, you should overlay the image and mask only the details you want to integrate and avoid those that makes the image look weird.
This is seriously great stuff—really love all the effort you put into this!
If you’re ever looking to try something different, GenYOU is great for such purposes (you can even use your own face for fun), and Ready Player Me is also worth checking out.
