You’ve seen AI-generated art, whether you know it or not. Marketers have been using it to generate ‘composite’ faces for years. You, know, that guy who looks like Tom Cruise with a little bit of Matt Damon mixed in? The combination seemed just a little too real.
Anyhow, those same tools are available to everybody today, and they’re providing a whole new way to start for a stencil image. I’m going to walk you through how to use MidJourney to generate great high-concept, low-res art that makes perfect stencils.
The image I want to generate is an image of Frida Kahlo. I like the Frida Kahlo stencils that are already in the catalog (like this one), but suppose I want to make one that is *not* based on a photograph. Then I can use MidJourney to generate the image, and then run it through Bay Stencil to get my stencil.
(So, if you’re following along, here’s a how-to for getting started with MidJourney.)
Below each image is the MidJourney prompt used to generate the image. And beside each MidJourney-generated image is the (best) Bay Stencil generated stencil from that image. You’ll see that some images from MidJourney are great, but then you see the stencilized version is less strong. We’re looking for prompts and results that look good *at the end* as a stencil, so we’re changing the prompt as we go along to get better results for stenciling.
I like the image, but the colors don’t suit: I was looking for a midtone painting that I could do with black, midtone and white.
I like the diamond-like stylized elements in addition to the flowers, but this doesn’t look anything like Frida Kahlo: instead it looks like a fashion illustration.
same prompt, but better results: it still has kind of a pout-y big-water-eyed fashion illustration look to it, but I like the combination of serious expression and flowers that I feel like I’ve seen in ‘good’ photographs of Frida Kahlo.
I really like this one, too, but more as an illustration than as a representation of the artist. I really liked the one before. Now I’m just editing Frida Kahlo, which I admit I don’t have any right to do–so let’s stop.
But so I want to go back and I want that image, I just don’t want it to be cropped at the top. I could bring it into Procreate on the iPad and try to illustrate the top of the hairdo, but it would be cooler if I could get an AI to finish the painting at the top, to ‘uncrop’ it…
Enter DALL-E 2, which has a new feature called ‘outpainting’.
I uploaded to DALL-E 2, and was able to get it to enlarge the canvas, and finish it off.
I’m just going to say right now that our A.I. friends have a bias for certain types of features, and together we have produced and ‘selected’ a Frida Kahlo that never existed. We probably shouldn’t say Frida Kahlo. But that’s what I typed into the prompt, so I’m going to leave it at that for now. I like this image, and encountered a lot of surprises while making it.
* DALL-E 2 is not to be confused with DALL-E mini (aka Craiyon), which totally sucks compared to the others I’ve mentioned.