AI art generated by VQGAN + CLIP


coldplay chris martin concert comic unreal engine


love peace joy kindness artstationHQ


Greta Thunberg climate change campaigner


leta ai in berlin


life architect


mott macdonald artstationHQ


algorithm, google, cloud, deep learning, machine learning
In this blue image above, the keywords were suggested by Leta (GPT-3), so it is effectively AI generating AI images!

Technical details

GANs (Generative Adversarial Networks) are systems where two neural networks are pitted against one another: a generator which synthesizes images or data, and a discriminator which scores how plausible the results are. The system feeds back on itself to incrementally improve its score.

CLIP (Contrastive Language-Image Pre-training) is a companion third neural network which finds images based on natural language descriptions, which are what’s initially fed into the VQGAN.

VQGAN: Vector Quantized Generative Adversarial Network
Released by Katherine Crowson @RiversHaveWings and Ryan Murdoch @advadnoun in Apr/2021.

CLIP: Contrastive Language-Image Pre-training
Released by OpenAI in Jan/2021.

Settings:
Defaults.
Seed: 42
Image size: 600×600

Platform:
https://huggingface.co/spaces/akhaliq/VQGAN_CLIP
Please do not over-use this platform, as it is a small instance with a queue already!