When ‘Creativity’ is set at 3 or higher, no matter how I alter my prompt, the woman’s face is essentially the same.
I’ve seen other users post their work, and it all looks like the same woman. It’s as if the ai model was trained on a singular woman’s face, and if the prompt is female-based, one can expect to see her over and over again.
It’s kind of a deal breaker : (
Am I missing something? Does anyone have anything to add to this? Is there any work around?
Yes, that’s Topaz Girl!
do the same with a picture full of people and enjoy 
And this is acceptable to users and the community? To Topaz Labs? …that everybody’s generations look almost the same?
It wouldn’t be that difficult to enhance the algorithm by adding like 5 more distinctly different women (and men), and then having the result randomly mix/morph the features of those women. I do it all the time in Midjourney, so I know it can be done.
But I got to tell you, I’m not going to be keeping my subscription going if all I can get is one women’s face. Like I said, that’s a dealbreaker. And if I had known ahead of time that this was the limit of Gigapixel’s functionality, I would never have subscribed.
One question: why would you want a random face that doesn’t represent the original person at all? Regardless if it’s always the same face or different ones?
I always use face recovery with those generative models as that really preserves the original face characteristics quite well (especially the so-called „creative“ algorithm which is nearly always more true to the original than the older one).
That’s a fair question. And you’re right, I really don’t want a ‘random’ face. What I want is to steer the facial characteristics in a unique, specific direction via the prompt…and the original image of course.
But when one raises the Creativity and allows the model to deviate from the original image, one would hope that the prompt wording would have more influence over the outcome. But I don’t find that to be the case. As the Creativity is raised, the model seems to become more and more fixed on ‘Topaz Girl’, regardless of the prompt.
(Unless I am ignorant of some special word or character, which may be the case)
When push comes to shove, I would rather have a random face than the same face over and over again…the same face that every other user is going to have. It makes it difficult to create a unique image. To argue the point to a ridiculous conclusion, imagine a world in advertising where every model in every ad looks the same.
At any rate, I notice that Redefine is ‘BETA’, and so I should be way more forgiving. I just hope that, as the ai model develops, Topaz Labs remedies this issue. Because it is an issue.
Thank you for your input,
R
Ah,OK.
Yes, with this goal in mind you unfortunately will always get features of TopazGirl™ in the current state of the software.
Hopefully with more training of the models this can get better.
