Hannah Patronoudis
Supernatural abilities, epic action scenes, and iconic characters that will forever stand the test of time: it’s hardly surprising that superheroes and their culture have captured our collective imagination for so long, enchanting both the young and the old. Be honest: at some point in your life, you’ve fancied which superpower you’d love to impress others with. But rather than always comparing yourself to established characters from Marvel or DC Comics, isn’t it high time to envision yourself in a superhero outfit, armed with nothing but your own creativity? Our creative technology studio, nocomputer, created a Superhero Generator that turns your ideas into reality with AI, all in classic comic book style.
Envision yourself as a superhero with our Generator in just two easy steps: take a cracking selfie and enter your desired superpower — which can be absolutely anything. Our Generator will then craft your superhero image. While it’s generating, you’ll get to read fun, GPT-generated snippets about the superpower you picked. For example, if you chose "football" as your superpower, don’t be surprised if one of your abilities includes "making perfect passes to your sidekick.”
In just a few moments, your superhero photo will be generated and ready for download!
But how does our Superhero Generator transform a photo and a word — or even a sentence — into a full-fledged superhero image with a specific theme? Let's review the several steps leading up to the final result.
First, we need a real superhero pose. We generate a pose with Stable Diffusion, and then it's time for some pose estimation—this is a computer vision technique that can pinpoint the key body joints of a person in images to understand their pose. We use OpenPose to create a stick figure in the same position.
In the next step, the superhero pose is fed into ControlNet, which ensures that the generated image matches the pose created by OpenPose. The face is masked during this step, so ControlNet works only on the body, leaving the face unaffected. For the face, we use FaceID and InstantID, tools that ensure the facial structure and hair closely resemble the selfie.
To achieve the characteristic comic book style, we then employ LoRA, which, in this case, is a fine-tuned version of Stable Diffusion specifically adapted for comic aesthetics. Finally, ReActor is used to perform a face swap, ensuring the face in the final output closely matches the selfie.
This whole process gets exported as an API on ComfyUI, showing the entire workflow clearly:
This sophisticated blend of tools results in a high-quality final image with your head on the body of a superhero. The world is not yet prepared for Broccoli Man, Captain Sleep Anywhere, and many other new superheroes - DC Comics, Marvel: beware!
Don’t get it twisted: the concept behind these superhero photos is merely one of the numerous potential applications of this technology. It’s enough to employ another fine-tuned model (LoRA) to take the end result in a whole new direction. Think of pirates, football players, House of the Dragon characters, or something entirely different.
Although this application isn’t in use anywhere just yet, we consider this Generator as an effective way to give people a quick taste of the potential of GenAI in a fun, casual setting. Imagine a touchscreen kiosk where you could generate images at fairs, festivals, or exhibitions—something like that could offer visitors a cool, lasting memento.
Interested in seeing the Superhero Generator in action? Reach out to us!
Written by
Hannah Patronoudis
Want to know more?
