AI Elphaba design renderings from different models

 In the ATCA brainstorming session about potential ways to use generative AI in costume class lessons and assignments (post forthcoming with a list of these), The idea was proposed to run the same prompt on different AI models. 

For comparison, I used the exact same prompt language from my first Copilot Elphaba with Google Gemini and ChatGPT. Here are the results.

Google Gemini 5 seconds to generate    ChatGPT1.5.hours to generateIt let me know that the reason the time to generate was so long was because of the number of people using the image generation tool at the time I submitted my prompt.   
 Microsoft Copilot15 seconds to generate

These outcomes are so fascinatingly different. I’m not sure how to interpret the results because the pathways these models take from input to output are so obscure. I assume it has a lot to do with the datasets on which they’ve been trained, and probably the prioritization among those datasets. 
It gives me pause that the platform my university has partnered with (Microsoft Copilot) is the one that provides the most objectified and cartoony output. Not passing judgment on that style, but if I’m using it in a fine art context, I’d prefer the output to be informed by a broader range of artistic styles, and a more precise execution when a specific style is mentioned in the prompt.
As an assignment, this would definitely offer interesting avenues of discussion and analysis. 
And on a personal aesthetic level, of these the one I like best is Google Gemini, so I decided to work with it a bit more. I revised my prompt to ask it to generate the full figure head to toe and here’s what it gave me:

This one took a bit more time at 13 seconds to render but nothing so long as the ChatGPT result, and I liked it enough to prompt it to render a design for Galinda in the same style and here’s what I got:

Another 13 second render Because Gemini quickly generated 3 different images, I asked it about image limits and here’s what it said:

Yes, there are limits to the number of images Gemini can generate, and these limits vary depending on whether you're using the free version or a paid Google AI plan (like Google AI Pro or Ultra).1

Here's a general overview:

Free Users: Typically have a more restricted daily limit, often around 10-20 image generations per day. There have been reports of users hitting limits around 50 images in a single chat on the free version.Google AI Pro/Ultra Users: Enjoy significantly higher limits, with some reports suggesting up to 100-150 daily image generations. They also get higher resolution and more detailed outputs, faster processing, and priority access during peak times.

Other factors that can influence your image generation capacity include:

Server load and current system demand: During high usage periods, you might experience temporary slowdowns or limits.2Complexity of your prompts: More complex image generation requests can sometimes consume more resources.Account history and usage patterns.Ongoing platform optimization efforts.

If you reach a limit, Gemini will usually notify you and tell you when your capacity will refresh, which is typically after a 24-hour period.3

It's also worth noting that some models within Gemini might have specific limitations, and the rules can evolve as Google continues to develop and optimize its AI services.4

I don’t teach an intro to costume design class, so this is probably the end of my experimentation with these AI models. I appreciate the opportunity to discuss this topic with my colleagues from schools all over the US and Canada. 
One more post on this topic is forthcoming, enumerating the assignment ideas, and applicability recommendations for those who teach at a university where the administration is strongly encouraging incorporation of the use of AI models in teaching.