The 3D characters I bash together from the turnaround angles look just fine in-game full body shots. But anything closer than that and the ugly truth begins to reveal itself. If these characters are to be used in any sort of closeups, I would need portrait images. So I set out to see how I could use Midjourney to generate a portrait image for the bearded weirdo.

First, I need to set up some close up cameras un Unity to get a reference mugshot.

As you can clearly see from these images, the character closeup is pretty horrid.

Now that I have this, it is time to head on over to Discord. I have set up Midjourney on my own Discord server – the channel is public so you can freely see all the generations I will be making for this game. I copied the seed of the original character turnaround and kept large parts of the prompt intact too. very old and weak man, cyberpunk point and click adventure game character, full beard, 2 ponytails at either side of the head, eyeglasses. portrait, rimlight –seed 2153840820

I actually had to change the prompt only once, to add the mention of the odd ponytails. The prompt produced following imagery:

Some of these portraits would make great characters in their own right, but only a few of them looked remotely like the 3D character. This is one of the biggest flaws of current generation AI image generation: the images, especially the details, are very random.

But in there I had one image I especially liked, so I generated a bunch of variations on that image:

I did not try to get a perfect portrait. Just get enough of them so that together they would make one that was good. Some of these had great beards, some had nice jackets, some had nice hair or eyeglasses. I could easily mix and match these in Photoshop for the perfect one.

I picked up some for high res versions:

In the end I ended up taking the head from one of these images and the jacket from another. So the final portrait is a combination of 2 generations.

Simple photoshop document

Here is the final image side by side with the in-game screenshot:

All in all it took me an hour to get to the final image from the 3D ref. Including writing this blog post.

One response to “AI assisted portrait generation”

  1. Avatar

    Awesome. Will be exciting when we have the ability to take an image and translate it back into 3D.

Leave a Reply

Your email address will not be published. Required fields are marked *

Recent Posts


Social Links