I use Leonardo AI for design.And it designs very good set of images. However a lot of those images have a lot of issues. Like it can produce bad images. Like no ears human, 3 hands and 1 feet human. And also some cases it can produce nudity too. So that kind of is a bad programming.
In such case you have to make changes to how you view the instructions in the prompt and how you get the design output. So for SDXL which is stable diffusion model, you have to give the negative prompt to the AI to generate the right set of the images.
Image Credit: Newscast
So what is our problem at hand? We get some of the bad images where the images are generated with the flaws.And in some cases we can ignore however in most of the cases that is flat out rejection and the bad quality of the art in general.
So how come the negative prompt is going to be helping us generate better images? In such case we have to think of the ways for the art to get better. In such case we make use of the negative prompt to generate proper images. And what type of the message we do give in the art for the prompt? Let's take an example below.
Take example of the image where we have a princess who is walking in the garden. So here's what we would be adding in the negative prompt. Check the below quote.
bad eyes, bad anatomy, bad figure, 3 hands, 1 feet, 6 fingers, three eyes, nude, three breasts, open body
That's just an example. But if you want the image to be more clear, you would require the AI to be trained on very small set of the details. And that is the only way you can generate better set of the images. Some way that can be done very clearly and it can even improve your life as well. But we don't know what could be the output. So we have to keep the iteration.
If your program allows negative prompt then use it. Make sure to regularly iterate and improve the workflow for those images. And also you make sure to design the images in such way that it can be used in prompt images. You have to experiment with the variety of objects or the art type you don't need and then make changes.
Hopefully you get an idea on what you need and don't need in an image to generate better quality through the SDXL.
