AI croquis: genial dad

There are many valid, ethical concerns about the use of these prompt-driven generative AI models—sustainability and ecological impact, inherent bigotry/bias, exploitative labor practices, violations of copyright and content use consent…the list goes on—and this post is not to enumerate or discuss those, although they would be excellent discussion topics to bring up with students in class. This post is aimed at educators, whose institutions are encouraging/requiring them to incorporate the technology into their teaching. 

In my ongoing experimentation with using my university's Microsoft Copilot AI assistant in costume design applications, my first attempt to create a design rendering for the character of Elphaba in Wicked was initially underwhelming and then maddeningly unacceptable.

 I decided that perhaps--especially in light of the datasets on which the model has been trained--jumping right into a fully developed costume design rendering was perhaps putting the cart before the horse, as it were. 

Instead, how about using it to create a croquis, a physical body sketch, that could be printed out or imported into a digital drawing app for adding the garments? Figure drawing is often the most challenging aspect of costume design rendering. I took a semester of life drawing with live models in the art department and still struggle with it.

Because my Elphaba results came out so objectified and idealized (like most media depictions of women, which probably figures significantly in the training data), I decided to try a masculine croquis render. I also thought perhaps specifying physical qualities like height/weight might yield a result that didn't look like a stylized superhero. 

Here's my first prompt:

create a fashion-design-style croquis body sketch  

in a pen-and-ink style for a 5'5" tall man  

weighing 200 pounds

Um, not what i intended. 

I was hoping that prompt would yield a regular-guy croquis, and instead the result is apparently a fitness model. Which, again, I'm sure the training dataset does include a lot of fitness models, and underwear models, and actual superheroes. 

Again I'm surprised but not. Again I see in retrospect how i erred.

At least this feels like a promising place to start from, given my desired outcome. So I wrote a revision prompt:

revise to have a flabby dad bod.  

he should be zaftig and his pose more playful

But Copilot had an issue with my language:

Note to self: Look up this Designer code of conduct. [1] 

Don't get me wrong, I'm glad that there are apparently language restrictions on prompt phraseology to (hopefully) keep users from creating hateful, mocking imagery, but I considered what language it flagged as problematic. I decided to resubmit the same prompt without the word "flabby." 

And it produced a charming revision:

I feel like I (or lets be real, a Costume Design Intro student) could start with this croquis and design a costume for a character evoked by this pose.

I still think there are overwhelming reasons to avoid using generative AI--sustainability,  bias, exploitative labor practices, etc.--and there are easier ways to get a croquis (like the MyBodyModel app), but if one's university is requiring it or mandating you find ways to incorporate it, this may be a path to explore?

I tell my students that if they're using a technology just for the sake of using it (for example, 3D printing a belt buckle design that would be more quickly and durably sculpted in Apoxie or polymer clay), that's a valid choice but it's not innovative. And this feels the same--it's a way to get a croquis but it's not an improved way.

 

[1] The Copilot Designer Code of Conduct is part of Microsoft's broader AI Services Code of Conduct and content policies. Here are the key principles that guide how Copilot Designer operates and what it allows:

✅ What It Encourages

• Responsible AI Use: Designs must align with ethical AI practices, including transparency, fairness, and safety.
• Respect for Individuals: Content should not misrepresent people or create misleading or harmful depictions.
• Creative Freedom Within Limits: Users are encouraged to explore artistic and design ideas, as long as they stay within respectful and safe boundaries.

🚫 What It Restricts

• Harmful or Objectionable Content: This includes violence, nudity, hate speech, or anything that could be considered offensive or inappropriate.
• Misleading or Deceptive Imagery: AI-generated images must not be used to impersonate real people or mislead viewers into thinking they are real.
• Sensitive or Stereotypical Depictions: Certain prompts may be blocked if they risk reinforcing stereotypes or include sensitive terms (e.g., "lumberjack" or "sword" have triggered blocks in some cases).

🛡️ Usage Requirements

• Implement safeguards to prevent misuse.
• Clearly disclose when content is AI-generated.
• Ensure human oversight and feedback mechanisms are in place.

You can read the full Microsoft AI Services Code of Conduct here

.