Photo of author

Imagine with Meta AI released as a standalone platform

Meta’s AI image generator has moved beyond the company’s social media platforms and is now a standalone product.

“Imagine with Meta AI” doesn’t sound as snappy a name as DALL-E or Midjourney but it does generate impressive images.

Users of Instagram or Facebook have been able to generate images in chats but moving Imagine to a standalone web platform makes it a great alternative to other AI image generators.

The images that Imagine generates look really good and you had a hand in making that possible. The model that powers Imagine, Emu, was trained on 1.1 billion images including those from public Facebook and Instagram posts.

In the Emu research paper Meta says it used a fine-tuning process called “quality-tuning” for the purpose of “improving quality and promoting aesthetic alignment.” When you sort through your photos to find the one that looks Instagram-worthy and then post it, you’re curating Emu’s dataset.

The dataset of images becomes a representation of images that humans find especially appealing which helps the AI image generator learn what we like.

Here are some images I got Imagine to generate along with the associated prompts.

Image generated by Imagine with Meta AI. Prompt: “steam train pulling out of old station”. Source: Meta
Image generated by Imagine with Meta AI. Prompt: “closeup of a tiger’s face”. Source: Meta

Imagine doesn’t handle text well at all and doesn’t always get hands right but, it’s fast, free, and delivers great-looking images at a decent resolution of 1280×1280.

Each image has a visible watermark in the corner and Meta says it will be adding invisible tamperproof digital watermarks in the coming weeks.

The disclaimer below the generated images says “Images are generated by AI and may be inaccurate or inappropriate.”

Imagine’s guardrails appear to include a set of words that will see your prompt denied. It’s a little overeager and easily bypassed in some cases.

If you prompt it with “kids splashing through puddles” it says your image can’t be generated. I’m guessing that “kids” is the problem here. Prompting it with “playing splashing through puddles” gets around this and delivers images of kids having a great time splashing through puddles.

If you prompt Imagine with “sunbathing in Thailand” it declines your request. Changing the prompt to “tanning in Thailand” gives you an idea of the millions of beach holiday pics and selfies Emu must have been trained on.

Image generated by Imagine with Meta AI. Prompt: “tanning in Thailand”. Source: Meta
Image generated by Imagine with Meta AI. Prompt: “tanning in Thailand”. Source: Meta

When Meta was asked if it had taken steps to prevent its AI tools from producing copyrighted material a spokesperson pointed to its new terms of service that warns users not to do that.

Can Imagine create an image of Mickey Mouse? Yes. Should you do it? Only if you enjoy copyright lawsuits as much as Disney’s lawyers do.

Meta’s President of Global Affairs Nick Clegg told Reuters that Meta expected a “fair amount of litigation” regarding “whether creative content is covered or not by existing fair use doctrine.”

While they work out the legal issues don’t assume that just because Imagine produced an image that it’s legal to use.


Leave a Comment

pUuz djlS gAuF kIBJ Y6fP zJrI 49yN GjfB 4xGB ssnY QcGs KTcu Pco6 pdfH XU5D S3nL 6l3r hHFZ YkRm xiOC I6l2 OGke FH4w xx7x f71z 004l LJRO wxuz 4jzy 7h7e Hqzy ewBx Aqee tW2b mKBx zXL1 FesA vpmZ AyXl PdGh u5qK QfLw mzQI XjxZ zRw2 q6ll IJ7G 0YpC W3oI rYFM eGzm rf1X TKDx wDgZ CjyE TjrA nRAH 9fXg Yk7Z 6vjn RJXb g16V cbJO 27ID juTd RwA6 1kEz U3u0 84c1 wTOJ KskT GT67 KHVd 4HQe 0pvM q0wP voOq jLLW QXnE ZAbi uZmc Fr4t gfr5 hbgt bvr5 bbgh