Type, and it shall appear

Text-to-image has reached the masses, y’all. Over the weekend, TikTok unveiled “AI Greenscreen,” an in-app, text-to-image generator allowing users to type out a prompt the software then turns into an image that can be used as a video background.

✍️🖼 For example… if you typed “POV reading this newsletter” into the filter, it’d probably pull up a picture of a person scrolling on the toilet – or someone sneaking in a quick news update in between firing off work emails.

But these images won’t be as photorealistic as the ones you may have seen from OpenAI’s DALL-E 2 or Google’s Imagen. Rather, they’ll look more like abstract blobs – and this is probably for a couple of reasons.

• The advanced models require a whole ton of computing power, which would end up being expensive and difficult to implement and maintain at scale.
• And speaking of scale, TikTok has over a billion users. Even if the company has unlimited resources, allowing everyone with the app to have unfettered access to create anything they wa… well… you get the picture. This is almost certainly why attempts to generate sexual or violent images in the filter currently fizzle out.

📈 Bottom line: In January 2021, OpenAI unveiled DALL-E, the world’s first-ever AI-powered text-to-image generator. Less than two years later, one-plus billion people have access to a watered-down version of the tech – and more innovation is sure to come.