'Terraforms' (2021) is a generative series that uses a custom combination of VQGAN + CLIP, and RGB optimization and depth mapping techniques.

The idea of this artwork was to start from the most 2D perspective possible — a pixelated, 2d image of a landscape — and to end the sequence on the complete opposite end of the spectrum, within a breathing 3D pixel landscape.

This is a process of feeding CLIP a series of pre-composed image prompts that were generated using CLIPIT and stylized with python text to image. The depth distortion is applied to the generated images created from those prompts.

The continuous zooming 'feedback' effect is created by providing clip a sequence of identical images to feed back to itself while generating the next frame.

Special thanks to DeKxi and R for their alchemy and being constant sources of knowledge.