Portrait studies from the 3d Scan library at bodiesinmotion.photo
People who know me know that drawing in essential to my creative process. Over the last couple years I have been using part of my morning drawing time (yes, with a coffee… or two, or three), to create input drawings to test my Bodies neural network, which I trained on a portion of my BodiesinMotion.photo library.
The idea behind this “AI tool” is that I train it to learn the correspondence between my drawing style and photographic representations of the human figure, in this case photography carefully lit and shot by me in the studio. Then, once trained, I can use it to dynamically ‘paint’ my drawings in the style of my photography. It is a wondrous interaction, and there is a magical space where I can draw very stylized or abstracted figures and the neural network infers some very interesting anatomical results, always beautifully lit and shaded. The images here are from my wall of Caffeinated Diversions, fifty of the most interesting results from these morning experiments. The grey line drawings are my hand-drawn inputs, the coloured images the output of my Bodies network.
Beyond just the final images though, a large part of the magic that has captivated me when developing and using these AI tools is seeing the final image emerge as I draw it. Here is compilation of timelapses from these drawing sessions:
Behind the scenes – prepping a piece for my “Artist+AI: Figures & Form” show opening next week at Somerset House. Suffice it to say, this will be a LARGE composition (22,000 x 17,000 pixels!). My drawing hand aches.
Show is free and runs from 19-23rd of June, so squeeze in a visit to Somerset House if you are near London.
click for larger
Drawing, painted using my Bodies neural network.
Here’s a jumble of head studies from yesterday. I was drawing from the Quickdraw page at BodiesinMotion.photo – 16 frames, zoomed into just the heads, randomly composed on the page as they appeared.
Here’s a timelapse of the sessions:
Another morning drawing session from the BiM Scan library.