Meta Unveils 'Imagine with Meta,' a Generative AI for Crafting Images from Text
In a bid to rival Google’s Gemini project, Meta has unveiled its own innovative AI-driven web tool called 'Imagine with Meta,' which transforms descriptive text into images. This tool bears similarities to other AI systems like OpenAI’s DALL-E, Midjourney, and Stable Diffusion. Utilizing Meta’s own image generation model, Emu, 'Imagine with Meta crafts high-resolution visuals based on typed prompts. Currently, users in the United States can access it free of charge and receive four images in response to each prompt.
Meta shared that feedback from users has been positive regarding the use of the 'imagine' feature within chat applications for crafting whimsical and imaginative content. In a recent blog post this morning, Meta announced they are broadening the horizons of 'imagine,' making it available for use directly on the web platform. While the chat feature encouraged a playful, conversational use of the image generation tool, the web-based offering focuses on standalone image creation at no cost.
The introduction of 'Imagine with Meta' follows previous controversies surrounding Meta's image-generating algorithms, which raises questions about possible safeguards in this new system to prevent similar issues. Although the tool was not available for external testing prior to the release, it will be carefully monitored as more users try their hand at it.
Looking ahead, Meta has committed to incorporating watermarks into the images created by 'Imagine with Meta,' aiming to add an extra layer of transparency and tracking capability in the following weeks. An initial visible watermark is already in place. The proposed watermarks are designed to be undetectable to the naked eye yet identifiable by a specially developed AI model. Meta has not yet disclosed if this detection model will be released publicly. They assert the watermark's resilience against various image alterations like resizing, color adjustments, and compression-enhancing security.
This concept of watermarking is not unprecedented within the realm of machine-generated artwork. Various firms like the French company Imatag have introduced watermarking solutions that persist through common image modifications. Companies like Steg.AI utilize AI models for robust watermark applications, and industry leaders like Microsoft and Google have adopted similar AI-based watermarking methods. Meanwhile, Shutterstock and Midjourney have agreed upon standards to signal AI-generated content.
With increasing concerns over the authenticity of AI-generated content, such as Deepfakes from conflicts and problematic AI-generated imagery, there is mounting pressure on tech companies to clearly label AI-generated works. This has been exemplified by new mandates from China's Cyberspace Administration seeking that creators of AI content indicate such origins without impinging upon user experience, as well as recent calls from U.S. lawmakers like Senator Kyrsten Sinema for enhanced transparency, advocating the use of watermarks for distinguishing AI creations.