1940s color palettes

In a recent AI training offered by my university, the advice was given to start prompts by defining a role for Copilot, giving it a task, and specifying the format you’d like the output to be produced in. So I’ve begun to write my prompts telling it, “I am a costume designer and you are my research assistant,” or assistant illustrator, or clerical assistant, etc., depending on what I’m prompting it to do.

Whatever the output I’m looking for, it often suggests that it could assist me further by putting together a color palette.

In a costume design context, this is often not a case of simply picking a few colors I like. If the play is set in a specific time & place, it can involve researching colors that appear in that context. 

It can be more complex than that, such as for example, a show concept like “Bollywood Jane Austen” where the setting is Regency England but filtered through 1990s Indian cinema, but this is not a post about a deep dive on thematic color palette philosophy.

I wanted to test generative AI on the kind of visual research a costume designer needs to do and see whether it could synthesize that research into a useful color palette.

As a test project, I used a show I actually designed the costumes for in 2012, and created a color palette for, It’s a Wonderful Life the Radio Play. The show takes place in 1946, and the premise is that the cast are voice actors performing It’s a Wonderful Life as a radio play in a sound studio.

At the time, I created a Pinterest board of my visual research that informed the color palette for the costumes of that show, which I shared with the rest of the creative team and refined as the design process proceeded. It was an excellent way to share visual research with creative team members who might be located anywhere in the country. 

I gather Pinterest is no longer a good platform for this kind of visual research, because of its inundation with AI slop. Unfortunate, because it was so useful and helpful in costume design.

Here’s a screenshot of a section of that IAWLTRP color palette inspo pinboard, which gives you a good idea of the broad scope of the visual research:

At the time, there was an online digital tool that would take images you uploaded or linked to, pull a selection of the colors, and generate a color palette grid: 

I’m sure this tool was powered by machine learning, in the same way that the early versions of the Pantone app would pull Pantone colors out of a photograph or a digital image of an artwork and allow you to assemble a color palette. 

When you compare this color palette grid with the visual research on the pinboard, you can see that it worked well. 

Now let’s take a look at the color palettes Microsoft Copilot provided when I asked it to do some visual research from 1946 and provide me with two color palettes for gendered fashion of the time.

.

They’re not awful, they’re actually pretty decent but no matter how many different ways I phrased the prompt, I couldn’t get it to integrate these into a single 25-swatch block. Not necessary, they’re fine like they are, I just would prefer the information in a different format that it refused to acknowledge and provide.

So I find this application acceptable, but not superior to prior ways of doing it with simpler and more controllable digital tools. I acknowledge that this is a matter of personal preference, and I can imagine a designer who felt differently and preferred to use Copilot for this purpose.

I decided to fiddle around with the Pantone app (which I’m sure has integrated AI capabilities now) and I suspect this might be a better way to work with color extraction and color palette creation. Take a look at the screenshot from a composite analysis of the two palettes Copilot generated and my original palette grid:

Full disclosure: I haven’t used the Pantone app in months and it has updated since I last did—this is a screenshot of me futzing around trying to alter the automatic color palette (the row at the top). 

I also have the Canva app, which has a color picker and palette creation capabilities. I did not experiment with that, however, because I find it to be a less intuitively structured app. I’m sure it would work fine.

Conclusion: It’s not that Microsoft Copilot is incapable of this kind of task, it’s that there are specialized tools designed for working with images and color control that allow for more finesse.