A resonant renaissance…. Overnight, the MobileCoin community has been playing with a chatbot that takes a string of text and passes it though an AI-art-generator, sharing the resulting works with the group. I was struck by the creativity of the AI, and also by the evolving set of human prompts, as we learned from each other. There was an interesting social experiment too; the algorithm incurs some real cost to run (9 cents/image), and the account started with 5 MOB in a free-for-all until it was spent. It is maintained now by donations, from the nouveau patrons of the AI arts. There is also a reward for spectacular outcomes; any artwork that fetches 5 or more likes gets a small cash bounty. I suspect that will not prove to be necessary, as curiosity and social commendation kept people up all night.
Details: In a Signal group, any message that starts with /Imagine will submit the subsequent text to the art bot. The bot is running VQGAN+CLIP (a powerful pairing of two distinct neural nets) on GPUs. VQGAN is a generative adversarial neural net (GAN) that takes an image input and generates a similar-looking variant. CLIP is a neural net that judges how well a text prompt matches a given image.
Leave a Reply