Google may the search engine leader, but it's following Bing when it comes to a controversial new feature. Coming close on the heels of Microsoft's inclusion of DALL-E 3 AI image generation in its rival search engine, Google is adding a similar function of its own, expanding access to AI image generators even more.
Users will be able to generate images using text prompts directly in Google search engine itself, and the option to use AI to generate images will appear in search results. Google insists that it will have strict content filtering to prevent misuse, but recent cases make me wonder.
For many, the Google search AI image generator will be their first experience of Google's efforts in this growing area. The tool is powered by Google Imagen, which hasn't yet been given a general release in a standalone form. Google's been developing the beta version of Imagen for some time but only a very small number of people have had access to the full version through the company's AI Test Kitchen. The general public has only had access to a couple of very specific and limited Google Imagen-powered tools: a city generator and a monster generator.
But a wider community will now be able to test Google Imagen's wider capabilities straight from the Google search engine. For now, the feature will be opt-in only via Google's Search Generative Experience (SGE) and only in English in the US. Those who choose to opt in to using SGE will be able to generate images from text prompts direct from the search menu. AI image generation will also appear as an option in image search results, with a message suggesting that users generate their own images if they can't find what they're looking for.
You can specify in the prompt the type of image you want, for example a photo, drawing or painting, and SGE will deliver four results. The user can then select which of the results they like and download it as a png file. So far, it all sounds very similar to Bing's incorporation of DALL-E image generation. Google insists that it's applying strict content filtering safeguards to prevent users from creating potentially offensive content.
That includes a block on the creation of images with photorealistic faces and prompts that mention the names of famous people. Access will be limited to over 18s initially, and images will contain watermarking and SynthID metadata that identifies them as AI-generated.
But despite the assurances, it's hard not to have some doubts. Bing has made similar reassurances, but 404 Media has reported about how users have quickly found ways around the safeguards to create, for example, image of cartoon characters flying planes into the Twin Towers. We've also seen how quickly people have broken the safeguards on Meta's AI stickers for FaceBook and Whatsapp, creating naked images of politicians and more.
Google recognises that the tool may not yet be perfect, which is why it's opt-in only. It's also including a feedback mechanism that can be used to report instances where the tool goes off the rails. It's not yet clear if it intends to eventually charge for the feature. Access to the Bing AI image generator is free via the 'Create' tab in its image search tool, but users have a limited number of 'boosts' to use for generating images and have to pay for more if they run out. DALL-E and Adobe also run credits systems for their AI image generators, while Midjourney sellls subscriptions.
Other new features in Google SGE include a conversational mode in Search through integration with Bard, again much like Bing's integration with Chat GPT. Users can also write drafts in SGE and customise the length and tone of the output. The new features are available to those who opt in to use SGE via Google Search Labs in the US. To learn more about using image generators, see our roundup of the best AI art tutorials.