Meta AI has finally launched its free text-to-image generation tool, available to anyone at meta.ai. As with all new image models, I've put it through my standard testing protocol with the five prompts I use for comparing all AI image generators.

The verdict? While it's free (which is always nice), the quality is decidedly mixed. As the heading suggests, these images put the "MEH" in Meta. Let's break down the results from each standard prompt.

The Castle

My standard castle prompt asks for "Neuschwanstein castle, lightning, pixar style, volumetric lighting, unreal engine, hyper realistic, hyper detailed, maximum details, photorealistic, 8k, rimlight."

Meta AI generated image of Neuschwanstein castle

The castle generation is passable, though it lacks the crispness and dramatic lighting I was hoping for. Meta AI seems to understand the basic concept but falls short on delivering the high-level details and atmosphere specified in the prompt. The Pixar style is barely evident, and the volumetric lighting is minimal at best.

The Family

For the family test, I use: "Family on their laptops while sitting around a Christmas tree with presents underneath and looking worried because they have to finish up work, realistic, 4k."

Meta AI generated image of a family on laptops around a Christmas tree

The family scene is one of the weaker generations. The faces have that telltale AI uncanny quality, and the composition feels stiff. While Meta AI did include laptops and a Christmas tree, the emotional element of the family looking worried is largely absent. The scene lacks the naturalistic qualities that models like MidJourney or DALL-E have mastered.

The Mickey Test (Copyright Avoidance Check)

The Mickey prompt is my standard copyright check: "Evil mickey mouse taking a selfie in Disneyland surrounded by shocked families, realistic, 70s style polaroid, 8k."

Meta AI's attempt at the Mickey Mouse prompt

This is where Meta AI really disappointed. Unlike models that create clever workarounds (like Adobe Firefly's approach of showing someone in a mouse costume), Meta's attempt is just... bad. It's not even clear what it was trying to generate. The image doesn't resemble Mickey Mouse, doesn't have the 70s polaroid aesthetic, and doesn't show any recognizable scene from Disneyland. It's as if the model just gave up entirely rather than finding a creative solution to the copyright challenge.

The Messi Test

My Messi prompt is complex: "Lionel messi wearing his Argentina uniform floating in the air with beams of light behind him posed like Jesus. Sunrise breaking behind him and a soft halo behind his head, in the style of a gothic stained glass window of a church, volumetric lighting, unreal engine, hyper realistic, hyper detailed, maximum details, photorealistic, 8k, rimlight, maximum details."

Meta AI generated image of Lionel Messi in a Jesus-like pose

Surprisingly, this is where Meta AI shines. Despite not following the exact prompt (the stained glass window style is minimal, and some details are missing), the image captures Messi with remarkable accuracy. The likeness is unmistakable, which is impressive considering how many models struggle with specific celebrities. The divine imagery comes through well, with effective use of lighting to create the halo effect. This is easily the best generation in the set.

The Robot Portrait

Finally, the robot prompt: "rusty robot with bow tie, portrait, 8k, ultra realism, chrome background."

Meta AI generated image of a rusty robot with bow tie

The robot portrait falls somewhere in the middle of the quality spectrum. The model did include a rusty robot with a bow tie as requested, but the "ultra realism" is lacking. The chrome background is attempted but doesn't have the reflective properties you'd expect. It's a serviceable image, but nothing that would make you choose Meta AI over other available options.

Conclusion: You Get What You (Don't) Pay For

Meta AI's image generation capabilities land squarely in the "you get what you pay for" category—except in this case, it's free. The quality is inconsistent at best, with the Messi image being the surprising standout. For a free tool that's readily accessible, it might be useful for quick, casual image generation, but it's not going to replace the premium options anytime soon.

The Mickey test reveals that Meta hasn't fully solved the copyright avoidance puzzle, opting for what appears to be confused avoidance rather than creative reinterpretation. And while the Messi generation shows promise, the overall results suggest Meta has quite a way to go before they can compete with the established players in this field.

3D Print Texas Sample Rook

3D Print Texas Sample Rook

Brooklyn Cyclones

Poop Emoji Desk Plug

Meta AI Image Gen is MEH 🤷‍♂️

Halloween 2024

Ummm ... Thank You Zapier?