My lazy attempt at AI art and why human craftsmanship will always matter
not me starting a whole substack just to reflect(aka rant) about AI art from a complete amateur point of view
I love art. But I don't consider myself an artist. Especially when it comes to drawing. I could actually say that not knowing how to draw (beyond the stick people level) is one of my personal frustrations in life. So, with AI image generation becoming more and more popular, the thought of being able to have the robots bring to life the ideias I have in my mind of drawings I would never be able to create on my own seem very appealing at first glance.
In the past few days, looking at many examples of people sharing their AI images versions in Studio Ghibli, Victorian or whatever other style, I was inspired to try my own hand at seeing how ChatGPT would perform with making my amateur artistry dreams come true.
A bit of context and disclaimers before we continue:
I am not a professional artist (we already covered this), nor am I in any other professional arts related field. This is all curious amateur-hour shenanigans.
This whole thing was brought out of simple curiosity about how easy and accurate creating AI art would be for someone with purely personal motivations.
I am not a pro at writing prompts (yet…?), but I've done my share of personal research on it and use it successfully for work and personal stuff on a daily basis (Also, I do have some basic data analytics background and experience, but that's a long story).
I did not spend copious amounts of time writing long prompts for this test. Again, I was lazy. Or, to put it in prettier wording: I wanted to measure how fast, easy and seamless the process could be for a complete amateur.
My view of AI tools is that they are tools. Amazing tools. And as tools, they are 1) not self-sufficient, 2) nor intrinsically good or bad. It all comes down on how you use it and what you use it for.
So, this was my lazy attempt at testing AI art generation.
First, I decided on my parameters:
Subject: Severance (I mean, we're all still processing that Season 2 finale and why not just ride the hype wave?)
Tool: ChatGPT
Style: Scott Pilgrim vs the World comics (did you know that bread makes you fat?)
Then, I chose a random picture of the Severance cast from a quick search on Google Images and uploaded it to ChatGPT:
See what I mean by lazy? I chose this picture because I liked it and it was one of the first ones I found. Mark's face is completely cut out but I don't care, I have faith in the robot's skills.
Then I opened ChatGPT to start my laidback digital drawing playtime.
First prompt: Create an image based on the image I'm uploading. Turn it to Scott Pilgrim vs the World style.
First(ish) result:
Honestly, my mind was blown by how a simple prompt delivered such a cool result. It is instantly recognizable as a Severance cast drawing in a Scott Pilgrim style.
I say this was the first(ish) result, because technically it was my second prompt. The very first drawing the robot gave me had the random floating hand issue AI is known for when generating images. So, the picture above is the one after I updated the prompt by saying: “Just take out the floating hand in the drawing".
Easy peasy… to a certain extent. Notice Irving's badge is also just floating with no lanyard, and the robot decided on its own to add a clock on the wall (maybe as a reference to the clock they do have on the tv show? Is my robot assistant a Severance fan too?), which didn't bother me either way.
Now we get to the part where the limitations of creating AI art started to show up.
So, we are two prompts in right now. I am excited about the cool result I had, how easy it was to fix the first issue and I didn't even notice the floating badge thing till later. I wanted to keep going and move on to creating a more “original” drawing.
This was my third prompt: Now, using the exact same characters in this drawing, make them the Sex Bob-Omb band. Starting from left to right, make them the: vocalist, bassist, guitarrist, drummer. The drummer should say "we are Innies! 1, 2, 3, 4!" and the drum kit should have Cold Harbour written in it.
The result:
Again, for a first result, I was still pretty stocked. It was funny to see that the robot decided to put them in casual clothing, but I get it, I didn't specify anything about what they should be wearing. I'll give that to the robot (I'm writing lazy prompts, can't expect much). But, I did say I wanted the drummer to be the one talking, so Irving shouldn't be getting part of the speech bubble.
No worries, it should all be an easy fix just like the floating hand issue. Right?
Now we are going to the fourth prompt, and here's where my frustrations began.
Fourth prompt: Update the image. Keep using the exact same characters in the previous drawing you generated. Keep them in business clothes. Only the drummer should say the "we are Innies! 1, 2, 3, 4!". Everything else stays like this last image.
The result:
The consistency is starting to get a bit shaky. It corrected the clothes, but changed the strokes, colors and shades (again, not a professional, maybe I'm using the wrong terminology, and for that I humbly apologize). Even if those changes aren't big, they are still there, so it's interesting to see that the robot struggles to keep consistency and adjust only the parts that I've told it to. And then, there is the obvious issue: it did the opposite of what I told it to do when it came to the dialogue bubble.
Let's try again. I mean, I still have faith in the robot. This was probably a minor misunderstanding.
Fifth prompt: update the image, just change the dialogue bubble to the drummer.
Result: I'm not even going to upload the picture, because it gave me the exact same thing. Only the characters’ eyes now were all a bit bigger. Okay then.
Let's start over. I'm losing a bit of patience, after all, this was supposed to be a quick and simple fix, and this whole thing was meant to be just a fun personal project. If it's taking me more than 20 minutes it's starting to feel like too much work.
I decided to upload that first initial drawing of the cast that I liked the result so much, even with the tiny inconsistencies (that I honestly still wasn't paying attention to).
Sixth prompt: Bring the characters' drawing back to this image. Make them the Sex Bob-Omb band. Starting from left to right, make them the: vocalist, bassist, guitarrist, drummer. Only the woman should say "we are Innies! 1, 2, 3, 4!" and the drum kit should have Cold Harbour written in it.
The result: Nothing. It changed nothing. I guess not even robots are fans of multiple client requests and reviews. And I guess the robot doesn't know how to differentiate gender in drawings either.
Let's go for one last prompt. One last chance of redemption for my robot artist.
Seven and last prompt: Update the image. Keep everything the same EXCEPT that the woman with the red hair needs to be the only one saying the "we are Innies! 1, 2, 3, 4!".
Result:
When I got this result I just laughed. It felt that by now ChatGPT was making up one of those “find the 7 errors” games for me to play.
I'm already 25-30 minutes in and have already established that it is just not worth my time to try to keep this going half-heartedly. If I wanted to actually create the vision I had in my mind I would have to start over and actually do a thorough research on image generation to then, in theory, write a meticulous prompt that wouldn't require so many changes.
In other words, I would have to become a craftsman at prompting for this specific goal.
Honestly, I am not that surprised my little experiment went sideways. Not only because my prompts were lazy, but because like I said in the beginning: AI tools are nothing more than tools. And tools are meant to be used by someone who can master said tool. And(!), the more that person has knowledge, experience and command of said tool, the better the results.
I remember one time I heard someone say that you know a great photographer if you give them a shitty camera and they still take great pictures with it. Or maybe you have also heard a similar concept like the one about pianists: you need hands to become a professional pianist, but having hands is not what makes someone a pianist. Or that, most of us learned how to use scissors, but it doesn't mean that, because we know how to use scissors that we can now all start cutting hair (I mean, you can, but stay away from my hair).
With so much hype and excitement going on around AI we tend to get lost in what we see on social media, the apparent quick masterpieces we see online. And everyone chanting that the AI era is here doesn't necessarily mean it's true. Or it does, it just doesn't mean what people think it means. That AI will replace human creativity for something better or equal. That it has never been easier to create, and you won't need to put effort in creating things. Maybe you won't… if what you want is to create a version of a copy of what's already out there.
That is why craftsmanship still matters. It will always matter. Craftsmanship never stopped being important and it will always be needed when our goal is to create something more than the bare-minimum.
I'm curious to see how artists will continue to use AI to push boundaries and make new and refreshing digital art. I'm also curious to see how other artists will continue to create new and amazing work without the crutch of any AI tools (or even any other digital tools).
And for those of us just using ChatGPT for our own random creative pursuits, I salute you. Don't spend to much effort on it though, unless your goal is to learn how to write advanced prompts. If your goal is to create something, go create, and only use the robot if it actually makes sense for what you are trying to achieve.
Ironically and unexpectedly, this little lazy experiment got me writing again, something I enjoy and haven't done in a while. So I'll take it. Let's see when I get the urge to word vomit like this again.
(Can you imagine if I now ended the post with: this was created using AI? LOL. Thankfully I believe the human imperfectioness of my writting is shining through in all its glory).
Bonus afterthought:
I am extremely curious to see how the NY Times lawsuit against OpenAI is going to turn out, and what this will mean for copyright issues concerning making a profit out of things created using ChatGPT. That is a whole other rabbit hole I think the majority of people is not thinking about. All we want is that profile pic drawing of our faces in Studio Ghibli style. I also wonder if Hayao Miyazaki is planning to sue OpenAI for butchering his work? (Still going to ask my robot artist for that profile pic drawing though, might become my profile icon here [already have], no shame).