Create original images from text with Pollinations.ai
How to use different parameters and artist's styles to create synthetic images and achieve the results you want
Pollinations.ai is a platform for synthetic media, that allows people to create art made by Artificial Intelligence. The site gathers different AI models, each model has its particular identity and gives different results, so it’s nice to explore the different models on the website to find which one is better for your needs.
With simple text, anyone can create unique, copyright-free images, to be used to illustrate articles, publications, as artworks or NFTs.
I use Pollinations to illustrate Envisioning.io “Weekly Recap” a publication in which we gather news about futures and emerging technology. I’ll show how I do it, and explain how you can use the platform’s parameters to achieve the result that you want.
I like to use the notebook called “Text-To-Image - CLIP-Guided VQGAN” because it is great to copy the styles of specific artists.
If you are new to pollinations.ai, please start by watching the introductory video on how to use the platform to create images :)
What each parameter does and how to tune then according to what you want:
This is the form you’ll have to fill up to create an Image using the model “Text-To-Image - CLIP-Guided VQGAN”, let’s go through each of these parameters, and how to use them.
Text Prompt:
Here you must write what you want to see in the image. These are the direct instructions for the AI. You can also write the style of a specific artist using “by” and the name of the artist, or write several inputs and allocate different importance to them, for instance:
Text prompt: “landscape by malika favre”
Text prompt: “landscape by malika favre: 1 | pastel colors: 1 | oil painting: 0.5”
Width and Height:
The size of the resulting image, in pixels.
Super Resolution:
This is an extra step added to increase the size and resolution of the resulting image, I suggest always leaving it on “yes” .
Image Prompt:
Add an image here if you want to give extra instructions about what your result image will look like. I think it’s always better to add an image if we want to have more control over the result. For this article I am using this image prompt: *taken from here.
Step Size:
This number determines how creative your result will be. How much it will differ from your image prompt and text prompt.
Higher numbers mean greater change:
Original image:
Text prompt: landscape by malika favre
Step size: 30
Text prompt: landscape by malika favre
Step Size: 90
Iterations:
This number determines how many unique images the AI will generate, each one will be created based on the one that came before, so the higher the number, the more distant the result will be from the prompt image. Also, the higher the number, the longer it will take to finish the run.
I like to use a lower number, like 200, so I have more control over the result, and to have the result appear faster.
Text prompt: landscape by malika favre
iterations: 200
Text prompt: andscape by malika favre
iterations: 700
These are some artists and illustrators you can use as styles to create images
Here you can see how we can generate different results according to artists’ styles. You can use any artist you want, these are just some examples to be used as inspiration. All images were made with the same as a seed image I posted above, the parameters were the following:
Text prompt: landscape by [NAME OF ARTIST]
With: 712
Height: 512
Super Resolution: yes
Step size: 30
Iterations: 200
Bijou Karman:
Liam Cobb:
Hokyoung Kim:
Malika Favre:
Julian Glander:
Jillian Tamaki:
Roger Dean:
Mondrian:
Moebius:
Jack Kirby:
Hilma af Klint:
Hildegard von Bingen:
Pollinations.ai is free and open-source, join our Discord to get involved and get help directly from the community.
 | A guest post by
|