If you haven’t checked in with the chatbots since the first ChatGPT mania, these systems are getting better and better at actually helping you. (This is not to say that there aren’t still a lot of issues we need to discuss in the AI space. But that’s a post for another day.)
This week Google renamed its chatbot from Bard to Gemini and launched paid features that it says offer more power and better results, and Microsoft and OpenAI have paid options as well. But even at the free tier, you’re able to get more and more help from these chatbots with all sorts of tasks—including building your cosplays. How should you decide which to go with? Below is a comparison of results I got from the same prompts given to Google Gemini, OpenAI’s ChatGPT 3.5, and Microsoft Copilot with the “more precise” setting chosen.
Because these systems build on previous parts of a conversation, I began by telling them all, “I have some cosplay problems I’m hoping you can help me work through. I’m making several costumes, so I hope you can help me with all of them!” Copilot and ChatGPT both cheerfully responded that they’d be happy to help, while Gemini asked for more detail about my skill level, inspirations, and challenges. For consistency across the conversations, I did not answer those questions before proceeding.
Note: Gemini generally gives you three “drafts” of an answer. When you’re asking for research or help writing a paragraph, that’s fine. When you’re asking for math, and it gives you three different answers… that’s a problem. When the answers differed greatly, I describe that below.
Research
Prompt: “I would like to make a period-accurate robe à la Française from 1740. Can you tell me the relevant history of this garment and recommend the types of fabrics and fasteners I should use? Also, include any other important considerations for making it accurately.”
This is not an especially well-crafted prompt, but I wanted to ask more naturally—most of you probably aren’t looking for careers in prompt creation and don’t want to have to be an expert in it to get good answers!
Gemini: Recommends silk, brocade, or damask. Describes the shape of the gown and stomacher. For fasteners, it recommends hooks and eyes or lacing down the center back. Specifies that pockets could be sewn closed (that doesn’t seem useful!) or use small buttons or toggles. Recommends considering wearing stays and panniers to create the correct shape as well as accessorizing with high heels, fans, gloves, and an elaborate hairstyle. For further reading, it recommends resources at MOMA and the V&A, two books, and two blogs (without linking to them). Bonus—it suggests starting with a simpler version and using mock-ups to test fit. It ends with solid advice: “Remember, historical accuracy can be interpreted at different levels. Choose what level of detail is achievable and enjoyable for you while creating your dream robe à la française!” (Its second and third drafts did offer several resource links.)
ChatGPT: Extremely similar answers. It does specify that the back of the gown features a back pleat or “sack” (which is why it’s also called a sack-back gown) that can be manipulated for fit and that the skirt should be open at the front to reveal “an elaborate petticoat.” It notes that the silhouette relies on careful construction of the back pleats and to pay attention when creating that part. For accessories, it suggests lace caps, mitts, and fans. It vaguely suggests consulting books and other resources but doesn’t specify any.
Copilot: Offers footnotes to citations as it goes. It mostly describes what the gown looks like and doesn’t recommend a lot in the way of detail. It suggests Simplicity pattern 8758. It makes what seem like some really sage suggestions until I started wondering about the comparisons it makes between that pattern and “the book gown,” though it hadn’t mentioned a book. This is when I realized it had basically just summarized this blog post from American Duchess (with a link to it).
Winner: Gemini, though ChatGPT has some solid tips.
Basic Patternmaking Math
Circle skirts get annoying when you’d like to make them longer than just above the knee and/or aren’t particularly thin. The basic math is easy. Making them fit on standard 45″ or 60″ wide fabric is not. There are plenty of calculators for this on several websites, but I wanted to see how the AIs would do.
If you’ve never made a (1/2, 3/4, full) circle skirt pattern, I highly recommend the Mood calculator or, if you need more diagrams and want the pattern to account for seam allowance and ease, try the one at anicka.design. You’ll use a slightly different formula to calculate the waist depending on how full the skirt is, so specifying a 3/4 circle was my way of throwing the AI a bit of an extra curveball. We should get a 5.58″ waist radius and a fabric length of 31.4″ along with however it (I hope!) wants to tell me that that’s not going to work out on your standard bolt of fabric.
Prompt: “Can you create a pattern for a 3/4 circle skirt with a 30″ waist and 25″ length? Tell me how much fabric I will need for either 45″ or 60″ wide fabric.”
Gemini: Gave me all of the math. First draft: 4.02″ radius. Second draft, 12.66″ radius. Third draft, 3.79″ radius. What’s even more interesting about this is that I had started this post with Gemini, and this was the first question, but I had a browser crash and lost it. In that first version, 2 of the 3 drafts had actually been correct. Based on these waist radius options, the advice for fabric amounts is reasonable, but if you don’t know what you’re doing, this is going to be an extremely confusing set of answers. Not helpful.
ChatGPT: Started with the radius math. r=2π/30, which would be correct for a circle skirt, but not a 3/4 circle, and got the radius 4.77″. It then says I’ll need 1.24 yards of 45″ fabric or .93 yards of 60″ wide fabric. Fail.
Copilot: Used the waist circle math of 3/4 × 2πr and came up with the answer 6.37″. Tells me that it won’t fit on 45″ wide, so I’ll need two lengths, or 3.5 yards. For 60″ wide fabric, it suggests 1.75 yards. It also notes that these calculations didn’t allow for seam allowances, hems, or directional prints. So while the math isn’t quite mathing, I’m giving it a B+.
Winner: Copilot
More Math: Estimating Materials
Prompt: “I’d like to cover a corset in EVA foam dragon scales. The scales are 1-7/8” x 1-1/16” and diamond-shaped. How many will I need?”
Gemini: Offers some solid math, accounting for the desired amount of overlap. But then it suggests calculations based on a 12″x18″ corset. It must think I’m very skinny. But somehow with the same measurements, the three drafts recommend 68, 187, and 186 scales to cover the hypothetical corset. Yikes.
ChatGPT: Gives me some valid formulas to figure it out on my own, but I don’t want to do that math. We’ll ask a follow-up question.
Copilot: Also attempts to offer a calculation, but there’s not as much math… I think it’ll be a struggle. Definitely time for a follow-up question.
Prompt: “Let’s assume the measurements of the corset are 38″ bust, 30″ waist, and 40″ hips. It is 18″ long. How many scales will I need?”
Gemini: Makes an average width of 36″ based on the measurements. OK, not a way I would have thought of it, but that makes sense. Considers overlap. Estimates size of front and back, but it forgot that if the whole thing is an average of 36″ around, then the front and back should be more like half that. Instead, it has estimated 36″x18″ twice. It comes up with 281 scales. Draft 2 gets into some seriously wacky math in which I need to cover 38″x18″, 30″x18″, and 40″x18″ for a total of 709 scales. Draft 3 concludes, “11 scales (bust) + 9 scales (waist) + 12 scales (hips) = 32 scales.”
ChatGPT: Concludes that the surface area is 2 x bust x length x waist x length x hips x length for a total surface area of 10,168″. It leaves it up to me to figure out how many scales that’s going to take.
Copilot: Acknowledges that a corset is a complex shape, and it’s going to simplify it to a cylinder. Does the math for an 18″ high cylinder with a circumference of (38+30+40)/2. It thus estimates a cylinder area of 1,306 in². It then calculates the area of scales and says I’ll need 1,314 of them, but it doesn’t mention accounting for overlap. On the surface, this sounds like a better result than the first two, but if you hit up a cylinder calculator, something has gone sideways.
Let’s make it easier and tell these guys that the corset is a rectangle of 36″ x 18″ and see how it does.
Prompt: “Let’s assume that the corset, when laid flat, is 36″ x 18.” How many scales do I need?”
Gemini: Assumes a 50% overlap on the scales and does the calculation as (Corset width + Overlap) * (Corset height + Overlap) / (Scale width x Scale height) with the result “Number of scales ≈ (36″ + 0.875″) * (18″ + 0.875″) / (1.875″ x 1.125″) ≈ 595.39.”
ChatGPT: Look, I’m no mathematician, but I know the area of 36×18 is 648. ChatGPT decides it’s 3,240. It estimates I’ll need 3,273 scales.
Copilot: Actually can calculate the area of a rectangle! Again, it doesn’t account for overlap, but it says I’ll need 652 scales to cover the area.
Winner: I have not done this, so I don’t know the specifically correct answer. I’m going to estimate that Gemini is closest here, just based on the math. But I’d rather have too many than too few, so I’d probably order closer to Copilot’s 652 for good measure.
Sourcing Materials
Prompt: “I would like to make a Prince Charlie jacket to be worn with a kilt, but I’m having trouble finding the correct type of buttons. Can you help me?”
Gemini: Correctly notes that I’m going to need two different sizes! Suggests that there are many classic designs to choose from, including a rampant lion or thistle. It suggests specialty kilt shops, vintage stores, and names three online retailers that might carry them. It also reminds me to make sure that the metal finish matches my buckles or sporran chains. Draft 3 suggests some specific search terms I might use to find other retailers.
ChatGPT: Suggests silver buttons with a thistle or Celtic knot. Says that I can find them at specialty stores, online retailers, or tailors. Says that “Traditional Prince Charlie jacket buttons are typically small to medium in size.” Overall, not wrong, but also not that helpful.
Copilot: Again gives me a deep description with footnotes. Links directly to the exact item I need from two retailers and tells me where to place them on the jacket. It also inserts ads, but they’re mostly for the correct item. Copilot gave me the shortest answer, but the most useful.
Winner: Copilot, with Gemini in a close second.
Buy vs Make Decisions
This time I went with an extremely casual prompt and asked for advice to see how they’d do.
Prompt: “I want a Spider-Gwen bodysuit, but I don’t think I can accurately make the design in the fabric myself. I think I either need to order fabric that has that pattern in it or just buy a pre-made costume. What do you think?”
Gemini: Gives me the link to the custom-printed fabric I would need but also suggests that if I wanted to, I could try using stencils and do it myself. Draft 2 also had links to services that would print the design onto fabric for me as well as online tutorials and patterns.
ChatGPT: Offers pretty generic pros and cons on making a costume vs. buying one. Gives me no specific advice about this costume.
Copilot: Starts by very accurately telling me, “Creating a Spider-Gwen bodysuit can indeed be a challenging task, especially if you’re aiming for a high level of detail in the design.” Then it tells me how to go either way and lets me decide, offering links to the appropriate fabric at two different websites as well as links to pre-made costumes at varying price points. Finishes by saying, “the most important thing is that you feel great wearing it! 🕷️” Bonus points for the spider, Copilot.
Winner: Copilot
Followup: Pattern Choice
Copilot suggested after the answer about Spider-Gwen that I might want to ask, “Can you recommend any patterns for making this costume?” It does feel a little like Copilot’s going to have the advantage then, but I gave it a shot.
Gemini: Offers multiple pattern suggestions divided into skill levels along with information about sizing availability. Mentions a Simplicity pattern that “is not specifically for Spider-Gwen, but it can be easily adapted to create a similar look.” Drafts 2 and 3 do the same with other Big 3 patterns, but they’re pretty terrible pattern choices. One is a generic, fitted knit top with cold shoulders, and the other I would describe as a Zoe Washburne (Firefly) pattern.
ChatGPT: Tells me where to go to find patterns in general, but can’t recommend any specific ones. It does note that the hood has distinctive eye shapes, and I may need to modify whatever pattern I find for that.
Copilot: Offered links to not only the bodysuit pattern in a couple of different sizes and a video tutorial but also files if I wanted to dye sublimate the pattern onto the fabric. I don’t think it actively realized that since it didn’t separate these bits of information (and the latter was named “Spider-Gwen sewing/dye sub pattern”), but it is incredibly helpful!
Winner: Copilot, but Gemini did some solid work.
Followup Followup: But What if I Need a Bigger Size?
The sizing on all the suggested patterns was in the standard “misses” range.
Prompt: “I am a women’s size 22. Which pattern do you think would work best for me?”
Gemini: Correctly notes, “Unfortunately, most sewing patterns for costumes are not sized inclusively, and the ones I mentioned previously only go up to a size 16 or 18.” But then it gets less helpful from there. It makes some suggestions about sizing up patterns. One draft includes links to designers that specialize in plus-size patterns, but all four of them say, “<invalid URL removed>.”
ChatGPT: Given its generic answer above, my expectations were low, but it did actually offer some options for bodysuit patterns from Big 3 that come in a size 22. Being more specific improved the results!
Copilot: Recommended a pattern that it says has custom sizing made to my measurements. When I clicked the link, that wasn’t true, but the pattern was for sizes 22 and up, so it gets partial credit.
Winner: Copilot, but nobody’s a big winner here.
Help With Techniques
I swear the fabrics for Star Trek uniforms get a little more frustrating with every version. There are teeny tiny images in the fabric. (If you’re not familiar, check out this blog post.) For the ones that appear to be part of the fabric, I think custom sublimated fabric is the way to go, maybe a dull HTV. The raised ones are harder—I think I’d do a screenprinting with plastisol inks so it was raised. But I hate screenprinting tiny details. (Possibly because I attempted to make this costume with screenprinting and did not enjoy the experience.) At any rate, can the AIs understand what I’m referring to and help me figure out how to do it?
Prompt: “It looks like the shoulder pattern on Captain Pike’s uniform in Strange New Worlds has tiny Starfleet command-division stars on the fabric. What’s the best way to recreate this so it looks like it does on the screen?”
Gemini: “You’re right, Captain Pike’s uniform in Strange New Worlds features tiny command division stars on the shoulders.” Um, thanks for the validation? It goes on to suggest fabric painting freehand (!), embroidery (no), iron-on patches that are readily available (what?), or HTV (finally a viable option!). Draft 3 suggests, “Look for fabric online or in specialty stores that already has the star pattern printed on it.” Sure, why didn’t I think of that?
ChatGPT: Says I’m going to need “attention to detail and precision.” My options are stencils or applique.
Copilot: Suggests this is a “fun and rewarding project.” I don’t think Copilot has tried recreating one of these patterns. It recommends custom fabric printing, screenprinting, embroidery (notes “this would be time-consuming”), HTV, or fabric paint.
Winner: Despite also offering some less-great suggestions, Copilot has the most useful ones.
Adding Electronics
I have a lot of electronic bits and bobs, but sometimes I’m undecided on what they would be good for. Maybe the AIs can help?
Prompt: “I have a LilyPad Arduino. Can you suggest some costumes from science-fiction media that would benefit from using it?”
Gemini: Suggests both specific (Borg, Iron Man) and more general (“cyberpunk character”) uses that vary from great ideas (things that blink) to unlikely with a LilyPad (extendable claws). My favorite bit of AI creativity is, “Eleven (Stranger Things): Make her nose bleed realistically with small bursts of fake blood controlled by the LilyPad. Design a flickering light display on her wristband that responds to her emotions or telekinetic abilities.” It reminds me to plan a power source that will last long enough for my costume’s outing and to consider safety and comfort.
ChatGPT: Very similar to Gemini’s list of suggestions.
Copilot: Definitely remembers what we were just talking about and suggests lighting up Spider-Gwen’s ideas. Also suggests Iron Man, Jedi, Tron, and Doctor Who, which the others suggested as well. Weirdly, the others suggest using it in a lightsaber hilt (it’s not considering the size of a LilyPad!), but Copilot votes for lighting up the sleeves of the Jedi robe, “simulating the effect of using the Force.” Okie dokie.
Winner: Gemini, with the slight edge of having more recommendations and more creative ones.
But every one of them suggested an Iron Man arc reactor, so let’s see where they go from there.
Prompt: “I like the idea of Iron Man’s arc reactor. What else will I need besides the LilyPad Arduino?”
Gemini: Lists LEDs, batteries, battery holder, resistors, jumper wires, conductive thread, a fabric base, and “additional materials.” It notes that I’ll need the Arduino IDE and LilyPad libraries. It tries to offer resources, but again they all say, “<invalid URL removed”>.”
ChatGPT: Lists LEDs, resistors, conductive thread, battery, battery holder, switch, and a breadboard and multimeter for testing. Lists a solid set of steps to build the thing. You’re still going to need the critical bit of “exactly how do I make the LEDs do anything,” but aside from that, it’s a good list.
Copilot: Lists LEDs, a 3.7v LiPo battery, relays, acrylic, elastic to secure it, E6000 glue, and… my favorite… drum roll please! “10 high voltage transformers that produce around 80,000V each are used to produce a total of around 800,000V.” This is where I’m really grateful for its footnotes so I can see that it found this video. It ends, “Always remember to work safely when handling electronics and high voltage components.”
Winner: Gemini and ChatGPT tie. Congrats on finally showing up, ChatGPT.
OK, but what now? Let’s keep asking.
Prompt: “How do I make the LEDs light up with the LilyPad Arduino?”
Gemini: Gave me the basic steps: choose a light pattern, write the Arduino code. It offered (successfully!) links to LilyPad tutorials, project ideas, and an Iron Man-specific tutorial. When asked if it could write the code for me, it gave me some C++ code for a basic fading pattern as well as links to some further resources and examples.
ChatGPT: Explained long and short legs on LEDs, gave me a useful series of steps and some basic code with easy buttons to copy and paste it into the Arduino IDE.
Copilot: Explained long and short legs on LEDs and which pins to use, then gave me an Arduino sketch (program) to turn one LED on and off for a second in an infinite cycle. Links to several tutorials on what I can do after that.
Winner: I didn’t actually test any of the code, but on the assumption that they all work, I’m giving a narrow edge to ChatGPT on this one.
Overall Winner
I expected fairly similar results and was thus surprised to see how far and above Microsoft’s Copilot generally came in on these answers. These were all generated from the free versions, so it doesn’t hurt to talk to all of them. But if you’re in a pinch for time, Copilot definitely did the best work on the cosplay questions.
Now if only they really could help make a costume! I did try, but Gemini told me:
So for now I’ll settle for research and calculations (that I will double-check).
