I’ve just finished a months-long project involving re-rendering previously created AI abstract images, using Topaz Gigapixel AI Redefine along with NVIDIA 4090 GPUs. The results are a huge set of very interesting and very usable textures and images for many different purposes!
What are “Abstract Extracts”-?
My 9 month project was over 3 years the making! This huge undertaking began in early 2022 with my first foray into generative imaging. Using a mobile app, I ended up generating a total of 1759 lo-res, abstract-style images using various prompts such as “architecture”, “city scene”, “farm scene”, “ocean cave”, “sailing ships”, “scenics”, “window view” and others. The “interesting” results were cute but only looked good viewed small on the phone screen. Despite the fun I was having with this new technology, before too long I gave up on this app and filed the results away on a hard drive…
More recently, Topaz introduced a feature in Gigapixel AI which they called “Redefine”, using its own set of generative models with adjustable Creativity and Texture. I gave this a try on my various digital and film-scan photos as well as newer AI generated images at high settings and was amazed at the fun new details being created by the models from these sources!
At this point in November 2024 I remembered the old abstract phone renders, tracked them down (I did not delete them after all!) and batch-processed ALL of them through Gigapixel – twice! Using a couple of borrowed Dell 3660 towers with NVIDIA 4090 GPUs, I rendered all of these source images at 300 ppi at both 4X and 6X, which gave different results even when the same source was accessed.
This process took several weeks to complete, at several minutes per render even with the high-end PCs. But even this was only the beginning! During the first several months of 2025 I literally opened each one of these images in Photoshop and scoured through the generated details found in each render at high magnification. I cropped and exported out (extracted) themed areas that looked interesting (like a photographer choosing the best composition), did some minor touchup as needed to reduce or eliminate some of the more obvious AI oddities, and roughly filed these crops under various categories (textures, city scenes, people, foliage, water, animals, etc.).
When this was finally done, I went through the images yet again and fine-tuned the categorization into seven overall topics (with sub-topics), split into 14 manageable final collections. Finally, after months of work, the result of all this is just over 6600 “abstract extracts” adding up to 48.13 gigabytes of images ready for your use!
What can you do with these?
As described above, all the heavy time- and resource-intensive generative rendering has been done in advance, providing you with a huge supply of creative pierces to incorporate into your own designs.
Use these sets as a starting point for:
Textures/overlays
Image-to-image
Standalone illustrations
Composites
Inspiration for traditional drawing/painting compositions
I’ve prepared PDFs showing thumbnails of every image in every collection. You can also download 78 full-size representative samples, gathered from each category and sub-category. View these at 100% and see the detail as demonstrated below!
