Our First Test of Meta’s AI Image Generator Was a Hilarious Flop

Meta released a standalone AI image generator on Wednesday that will convert your text-based queries into fantastical, high-resolution images in mere seconds. The free tool, Imagine with Meta AI, is powered by Emu, the company’s image foundation model.

Meta rolled out this product to compete with OpenAI’s Dall-E, Midjourney, and Stable Diffusion. Imagine functions similarly to other image generators, but was previously only available on the company’s messaging platforms. AI-generated images from Meta come with a visible watermark in the bottom left corner and an invisible watermark that’s only perceptible by a corresponding AI model. The watermarks hope to add traceability and transparency to Meta AI, while also making Imagine’s creations resistant to traditional image manipulations like cropping, color change, and screenshots,

Imagine is free to use, and generates four images per prompt. Meta is rolling out its AI image generation models on the same day Google introduced the world to Gemini, its most advanced AI offering to date. Google and Meta are both trying to catch up with OpenAI and Microsoft’s ChatGPT, which dominated the world of AI for consumers in 2023.

So far the results from Imagine with Meta AI aren’t so impressive. Several of my attempts garbled my text query and gave me something confusing, and not what I asked for. Compared to Dall-E, Midjourney, and even smaller competitors like Playground AI, Meta’s product is severely limited. However, the service is free and unlimited to use, which cannot be said for all the others.

Last week, researchers from Hugging Face and Carnegie Mellon showed how terrible AI image generators are for the environment. Research showed generating one AI image uses as much energy as charging a smartphone. However, more powerful generative AI is what consumers want, and it’s going to take a lot of energy to win a race between Meta, Google, and Microsoft.