Vector AI Generation from text - Really poor basic results, after it was great for a few weeks?

[Eos128]

Hi,
I started a project using Illustrator's new AI generation from text to create some cartoonish characters, and the results were great. About 2-3 weeks ago I even noticed an upgrade where using maximum complexity in the settings I was able to create some really great looking characters that I was so happy with.

Today I came back to generate some more characters and even with maximum complexity, the results are extremely poorly and basic? So much so that the complexity setting even barely seems to make a difference...

The continuity of my project really relies on being able to get consistent results, and now the previous quality had set a really high standard... I really can't work with the new results I'm getting, and I don't know what's wrong? Is it a bug on your end? Please tell me this can be solved and we won't be getting these basic, inferior results permanently... I'm really worried about this.

[Eos128]

Also if it's about performance or something, please let us be able to get results with the previous quality, you can make it optional. You can see from the images I attached that a downgrade like this is ridiculous...

[Monika Gause]

Did you also rate them in the Text to Vector panel?

And then please do also post in https://illustrator.uservoice.com to reach the engineers.

[Eos128]

The quality was very consistent before for a given prompt, and now the exact same prompt just produces really bad results, also consistently... Something changed on AI's end somehow.
I've submitted an issue there, thanks for the guidance, hopefully it can get worked out.

[Bobby Henderson]

While the Text to Vector feature is kind of fun to play with, I could only use it in the process of brainstorming ideas. The results that are generated tend to be so random that it almost feels like playing a slot machine. I'm not going to use any of the results as finished quality artwork.

Any project involving a cartoon character used across a multitude of images might need a character system developed. That would be a series of heads with different emotional reactions and a series of full body poses to use as a reference to keep the look consistent across different images. Artificial Intelligence-based effects filters can't do that. Such things have to be done with human hands. A computer might need Artificial General Intelligence to accomplish such a task accurately. Computers with AGI capabilities would open a whole new frontier of possibilities and nightmares.