I've been using the amazing Snowpixel App for a few months now and I find it very user friendly for creators or just anyone looking to become familiar with AI and all the capabilities that exist to produce different types of art.
I've tried all of the various models and the Creative model is by far my favourite for producing clean images i.e faces that are mostly very symmetric. This did not happen right away. It took me a while to learn how AI works and I discovered a couple of things.
- The initial 10 images Snowpixel produces are really showing the various art styles that AI can produce for your prompt. They may not be very coherent but the key is to find an image that you like and then run the Similar model on it using the Low Similarity setting. With the new upgrade this will produce 50 similar images! (Personally think this upgrade is fantastic!). If you use the Medium Similarity there is a higher chance of you getting a lot of images that almost look the same as the source image.
- When picking an image for creating more images, I always try to find one that has pretty symmetric features or a feature that stands out like, nice eyes but the rest of the face may not be symmetric (and that doesn't matter). Sometimes the art surrounding the image is amazing and I want to create images with that type of background. Then I run the Similar/Low model I get some amazing portraits!
- Once you have your images, you will find several that are symmetric and you can evolve those further by using the Similar model and set the sensitivity to help achieve the similarity you want.
- Most of the time I find AI always has this 1 random image that it produces that has absolutely nothing to do with the prompt but tends to be pretty cool and you can make more of those images if you like. I always think of this as a bit of a gift. :-)
I have to say I've invested a lot in figuring this out but it was so worth it.
Here are a couple of examples.
As you can see the 3 images created from the first image are symmetric and are actually pretty beautiful!
Here is another example with some great faces I got from the prompt, "Merlin the Great Sorcerer."
None of the images were particularly coherent or symmetrical and then there was this surprise image:
From this old guy I ran the Similar/Low model and got these creepy but cool looking guys.
So in summary, I would say use a generated image as your source image and evolve from there to get the more symmetric portrait that you are trying to achieve. You can achieve the same results by starting in Draft mode and once you get your images continue to evolve them using the Similar/Low modes.
Hope this helps and have fun!
Are you making portraits with the Snowpixel AI? Tag us on Instagram @snowpixelapp so we can share your favorites!