Superhero Madness!

One of the assignments for this week was from Middlebury College and we were tasked with creating a strange picture using two different AI chatbots and compare our results. I took this assignment step-by-step and have three different versions of each step. So, my initial step was to just asked the two chatbots, DALLE-3 from OpenAI and SDXL Turbo from StabilityAI to give me a picture of a superhero. I was very open-ended with my question here and as we can see I got two very different results. On the left hand side, I got an image of a superhero that appears to have three legs and two capes and a mask that looks like the one Star-Lord wears in the Marvel franchise. It is hard to tell if this is where the inspiration for the image was drawn from because the colors of the suit and mask are very similar to that of Star-Lord. On the right hand side, we see a more typical superhero costume and an image that resembles that of Superman. We can see the differences from the two different AI chatbots and how one (OpenAI) is more realistic in terms of the picture quality and then the SDXL Turbo is more cartoonish when it comes to the picture quality. Moving on, I decided to add some more details to the prompts to try and figure out what the chatbots would output.

The next step of the design challenge was to add some flair to the images and see what you could create with the different chatbots that you chose. So, I decided to ask the chatbots to give me an image of each of the heroes that they generated in my previous prompt and have them standing on top of a rock with one leg being shown to be higher than the other like some heroes stand in movies. On the left, the image was generated by DALLE-3 and as you can see, it was pretty accurate with what I asked for and what I got. The image on the right is a little more obscure because it did not have the hero have one leg higher than another, however, it did have them standing on a rock facing the world and it also appears to have added some extra people in the back even though I did not ask for it. Also, the face on the superhero on the right looks to be a little more obscure as well and does not seem to be fully generated. Finally, I made one last request to the chatbots and that will be shown and explained down below.

Here the prompt for the chatbots was to create an image that had a superhero riding a cat on the moon. As a result, on the left, we can see that ChatGPT did a great job of creating the image that I requested and looks very nice overall. However, on the right hand side we see the SDXL Turbo output which was also tasked with the same job which was to create a picture of a superhero riding a cat over the moon. For some reason, this chatbot decided to ignore the “on the moon” part of my prompt and instead have the superhero ride the car in some random city. Also, this chatbot also decided to create a superhero that was a mix of both superman and batman and not just use the same costume throughout so that was interesting to see as well. Overall, the quality of the SDXL Turbo pictures were much worse than that of ChatGPT’s and ChatGPT was also able to be more accurate with its output and they were more appealing to look at.

I would say that the image on the right hand side was the worst image I got because it was not really that close to what I was asking for. I asked for a superhero riding a cat on the moon and I got a man dressed up as two different superheroes riding a cat in some random city. That is what I think that one was the worst because the output that I got was not what I really wanted at all. I would describe what I got as a child taking in instructions and only paying attention to half of the instructions. On a more positive note, the picture below I thought was the most impressive. I really liked how this one ended up turning out. I just think my simplicity of the prompt honestly made it easier for ChatGPT to interpret and output an image that was exactly what I wanted. Also, the sun in the background that is setting and the cape flying in the back is all extra from what I asked for and it just comes together to create the perfect image. One of the reflection questions in the assignment description talked about how AI may pose a threat to artists and where I think it would affect them. I think that AI poses a pretty serious threat to artists because whatever artists are creating, AI can create the same thing, if not better, in almost no time while also making it accessible to everyone immediately. Therefore, I think that AI will definitely pose a threat in the future because AI will take that sense of creativity away from artists because whatever they make, AI can also make. When creativity is lost, it is very hard to make interesting content so it makes sense that artists might struggle in the future.


  1. I would name the first one Tri-Ped, and say he kicks 50% more ass.
    The images tell me something about what is statistically correlated with the word “superhero.” I wonder if it’s possible to get an image generator to come up with something that doesn’t look like it’s trying to be a copy.
    There is a Youtube series, Strip Panel Naked ( that takes a deep dive into how graphic storytelling works. I wonder what it would take for an image generator to approach that level of complexity. There are massive datasets online to work with, but how would a Large Comic Model decipher and decode them? What kind of training parameters would it need?
    In addition to being a threat, I wonder if AI could be useful as a tool. If someone built an app that let an artist have some back-and-forth conversation to refine an image, and maintain some consistency across images, it could help a creator be more efficient.

  2. Pingback:Quick Turnaround for Weekly Summary #3!!! – Creative Bytes: Exploring the World of DS106

  3. Marisol Juarez

    First off, I wanna say I love how you work turned out! I enjoyed reading this post because you were honest on how the assignment was and how you compared both Chatgpt and SDXL Turbo giving you two different results. I found it a little silly that the SDXL Turbo complete ignore your “on the moon part”. As Well as, adding a batman mask on superman. This was such a good detailed assignment you did!

  4. Pingback:So, what did we learn? |

Leave a Comment

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