Meta launches independent AI image generator "Imagine with Meta"

After Google launched Gemini, Meta also launched a new, independent artificial intelligence generation experience on the Internet - "Imagine with Meta" (Imagine with Meta), allowing users to create images through natural language descriptions.

Similar to OpenAI’s DALL-E, Midjourney and StableDiffusion, ImaginewithMeta is powered by Meta’s existing Emu image generation model to create high-resolution images based on text prompts. It's free for US users (at least for now) and generates four images per prompt.

"We're excited to hear from people about how they're using MetaAI's text-to-image generation feature, imagine, to make fun and creative content in chat. Today, we're expanding how imagine can be used outside of chat," Meta wrote in a blog post published this morning. "While our messaging experience is designed for more fun back-and-forth interactions, you can now create free images on the web, too.

Now, Meta's image generation tools have landed the company in hot water of late (such as Meta being accused of offering a racially biased AI sticker generator), raising questions about whether Imagine with Meta has safeguards in place to prevent history from repeating itself.

Watermarks aren't available from the start, but Meta promises to start adding watermarks to content generated by ImaginewithMeta in the coming weeks to "increase transparency and traceability." Meta said these watermarks are invisible and will be generated by artificial intelligence models and detected by corresponding models. There's no word yet on whether the detection model will be made public at some point.

"[Watermark] is resistant to common image manipulations such as cropping, resizing, color changes (brightness, contrast, etc.), screenshots, image compression, noise, sticker overlays, etc.," Meta said in the post. "Our goal is to implement invisible watermark technology into many of our products with AI-generated images in the future."

Generative art watermarking technology is not new. French startup Imatag offers a watermarking tool that claims to be unaffected by resizing, cropping, editing or compressing images. Another company, Steg.AI, uses artificial intelligence models to apply watermarks that can withstand resizing and other editing. Microsoft and Google have adopted AI-based watermarking standards and technology, while elsewhere Shutterstock and Midjourney have agreed to guidelines for embedding markup to indicate that their content was created by AI-generated tools.

But pressure is mounting on tech companies to make it clearer that works are generated by AI - especially given the proliferation of deepfakes in the Gaza war and AI-generated child abuse images that bypass filters.

Recently, the Cyberspace Administration of China issued regulations requiring artificial intelligence generator manufacturers to label the generated content (including text and image generators) without affecting user use. During a recent U.S. Senate committee hearing, Senator Kyrsten Sinema emphasized the need for transparency in generative AI, including the use of watermarks.