experiments-ai – Kai Tattersall

GauGAN Stop Motion

GauGAN Stop Motion

Animating an environment into a face.

──❀*̥˚──◌─ ≪ °✾° ≫ ─◌──❀*̥˚─

NVIDIA’s GauGAN translates segmentation maps into photorealistic landscapes. The process allows users to delineate natural elements like “dirt,” “grass,” “water,” and “sky” and draw them in. It also allows custom map uploads.

I used After Effects to convert the sample footage (David Hoffman’s “1967 - Teenage Runaways Lost On The Streets of NYC”) into a segmentation map. Each frame was manually uploaded and processed.

Other stylization techniques were added for aesthetics.

Face Morph + Lip Sync

Making a big sing-along.

.•°¤*(¯`★´¯)*¤° . °¤*(¯´★`¯)*¤°•.

Artificial intelligence allows infinite generations of original faces. Combined with A.I. lip sync technology, I made a morphing collage of faces singing along to “Hand of God” by Nick Cave & The Bad Seeds & Warren Ellis.

The first section is a demonstration of the lip sync technology by itself; the scond incorporates face morph; the third is a re-interpolation of the third.

This was later repurposed for “Hand of God,” a more recent project.

Liquid GAN

Using artificial intelligence to dance.

꧁꧂꧁꧂꧁꧂꧁꧂꧁꧂

'
The relatively new “Liquid GAN” can sample human motion and apply it to another image. The process, at the time of completion, produced messy results. The limbs would disappear and the resolution suffered. I decided to lean into it.

FKA Twigs’ “Cellophane” felt like a good match. With “The Dance of the Sugar Plum Fairy” as a reference, six dancing models were created. I found that the starting frame affected the fidelity of the results most.

The background was created by translating club footage (found under CC on YouTube) into a segmentation map, then mapping it onto a landscape. Light beams became mountains, the stage--a lake, people--grass, etc.

In the months after, Liquid GAN improved enough to create a convincing model, and was used to creating a dancing crowd for “Lure of the Underground,” a more recent project.