Generate abstract paintings in one click
Click the button below for an AI-generated abstract painting. Built for artists, developers, and hobbyists.
It isn’t magic – it’s science. 1SecondPainting is a deep neural network trained on thousands of abstract paintings, including a number of pieces by Pollock, Delaunay, and more. This deep understanding of abstract art styles allows it to generate dozens of novel paintings in under a second.
Step 1. data discovery & analysis
Over 14,000 unique abstract paintings were scraped and compiled into a sizeable corpus.
step 2. cleaning & Formatting
The paintings were cleaned, pruned, and formatted according to best practices.
step 3. Training
A StyleGAN2 network was used to train 1SecondPainting through transfer learning.
step 4. fine-tuning & Generation
After training, the network designs unique, creative abstract art in less than one second.
I created 1SecondPainting (beta openly available until October) as a proof-of-concept to assess the utility of modern Generative Adversarial Algorithms in art and design (inspired by Gwern’s TWDNE). Technically speaking, the generator implements a pretrained StyleGAN2 network (FFHQ-512) fine-tuned through transfer learning on a corpus of 14,305 abstract paintings. On a Tesla T4, it takes on average 0.173 seconds per novel generation. Not bad!
You can find the model here if you’d like to play around with it.
Have a comment or a question? Reach me at firstname.lastname@example.org or on Twitter at @nicksaraev.