Google has been extremely careful about releasing its text-to-image artificial intelligence system, and while the company’s Imagen model produces an output of comparable quality to OpenAI’s DALL-E 2 or Stability AI’s Stable Diffusion, Google has not previously made the system available to the public.
Today, however, the search giant announced that it will add Imagen to its AI Test Kitchen app in a very limited form as a way to gather early feedback on the technology.
AI Test Kitchen was launched earlier this year to test a variety of Google’s AI systems. Currently, the app offers a number of different ways to interact with Google’s text model LaMDA, and the company will soon add similarly restricted Imagen requests as part of what it calls a “Season 2” update to the app. In short, there will be two ways to interact with Imagen: Urban Dreamer and Wobble.
In City Dreamer, users can ask models to generate city elements designed around a theme of their choice, such as pumpkins, denim, etc. Imagen creates sample buildings and plots (city squares, apartment buildings, airports, etc.), all designed to appear as isometric models similar to those seen in SimCity.
In Wobble, you can create a small monster. You can choose its material (clay, felt, marzipan, rubber) and dress it in the clothes of your choice. The model will generate your monster, give it a name, and then you can poke it and make it “dance”.
The “Wobble” feature lets users design a monster and make it dance. IT House has learned that these interactions are very limited compared to other text-to-image modes, and users can’t just create what they like. However, that’s what Google is intentionally doing. Josh Woodward, Google’s senior director of product management, explained that the whole point of AI Test Kitchen is to: a) get public feedback on these AI systems; and b) test which behaviors will make the system crash.
The big question, though, is whether Google will want to take these models to the broader public, and what form will that take? With the company’s competitors OpenAI and Stability AI currently rushing to commercialize text-image models, will Google feel its systems are safe enough to get out of the AI Test Kitchen and offer them to users?
