AI costume design, Part 3

This is the third post in a series about experimenting with generative AI to create a costume design rendering of the character of Elphaba from Wicked. 

Read Part One and Part Two here

At the end of part two, I hit my image generation limit for the day and could not revise any of the first round of artwork it created until the next day, when my limit would roll over. I thought that might be a blessing in disguise because I could really revise my prompt to include much more specific language to describe what I had in mind. 

Here’s my second prompt:

Revise the first image to render the entire figure. Thedress must cover the shoulders with a high collar. The witch’s body is muscularand sturdy. The dress is trimmed with leaves and vines. Her hair is in many longbraids in shades of green, gold, and brown. Her witch hat is rustic andangular. The sketch is rendered in the style of Alphonse Mucha

And here’s the result I got back: 

Now I’m not only disappointed, I’m pissed off. Not only has Copilot ignored many of my directions, this round is even more naked and impossibly thin than the first.

So, I told Copilot how I feel:

wow that's terrible.  

You ignored most of my requests for changes.  

I will draw it myself.

...to which it replied:

 Thanks for the honest feedback — I really appreciate it. It sounds like you have a strong vision, and drawing it yourself might be the best way to bring that to life exactly as you imagine.

If you’d like, I can still help by creating a reference sheet, pose guide, or even a color palette based on your description to support your drawing process. Would any of that be helpful?

I’m giving up on using Copilot to create a costume design sketch from scratch, and I don't see how it's useful providing any of the stuff it suggested--I don't need AI to choose my color palette, I just need my own opinion.
I think next, I’m going to experiment with using it to create a masculine figure. 
To be continued…