DALL-E Launched

We open January 2021 with a bang. This consequential deep learning model designed by OpenAI “generates”—as in creates—digital images from natural language descriptions. Utilizing a 12-billion parameter version of the GPT-3 neural network, DALL-E marked a significant advancement in AI-driven generativity. This launch opened the text-to-image race that continues today between DALL-E, Stable Diffusion and Midjourney. Although introduced at the very beginning of 2021, DALL-E and text-to-image generation did not receive mainstream attention until nearly a year later. This is because DALL-E 1 generated images autoregressively from discrete image tokens, which was computationally expensive and slow. It wasn’t until almost exactly a year later, in late December 2021, that the concept of latent diffusion models was introduced, which significantly reduced the computational requirements and decreased costs. This advancement would light the fire of mass adoption of the emerging technology.