Now Google’s AI-powered search Can Generate Images

Google has introduced the Search Generative Experience (SGE), which allows users to generate images from text prompts, starting on Thursday. Microsoft has already offered a similar feature using OpenAI’s DALL-E model through Bing Chat since March, making Google’s move somewhat expected.

Here’s how SGE works: If you’re part of Google’s Search Labs program, you can input your query into the Google search bar. SGE can then create several images based on your prompt for you to choose from. This tool is powered by Google’s Imagen family of AI models, according to Google spokesperson Craig Ewer.

In a GIF, Google demonstrated how the tool can be used to create images based on prompts, such as “draw an image of a capybara wearing a chef’s hat and cooking breakfast.” The GIF also shows that you can refine the queries used by Google to generate each image.

You can generate AI-generated images from Google Images as well. While browsing Google Images results, you may come across a prompt to generate images based on your query, and the generated images will appear in a sidebar.

Google emphasizes its responsibility in rolling out image-generation tools in a blog post. The tool is designed to comply with Google’s prohibited use policy for generative AI, ensuring it doesn’t generate inappropriate images. Google will also label images with metadata and embedded watermarks to indicate they were created by AI.

During the initial stage, Google will not allow the generation of images depicting photorealistic faces or prompts featuring notable individuals. Google is currently limiting image generation to users aged 18 and older.

Additionally, Google is adding a feature to generate written drafts directly from the search bar using SGE. You can request drafts based on your query, and the tools can create drafts of varying length and tone. If you’re satisfied with the result, you can export the draft to Google Docs or Gmail.

As for the models behind the draft writing feature, Google is currently using a range of Language Model Models (LLMs) trained on publicly available data sources and human supervision data. Since its launch in May, Google has been steadily improving SGE with more video content and better links, making these new image and draft creation tools potentially useful, although some users may still prefer traditional search results.

Latest articles