Meta's AI Image Generator Review (Yes, There's Another One)
There haven't been any significant developments in the AI image generation space since DALL-E 3 a few months ago — until Meta suddenly introduced its latest AI image generator. In this article, I'll discuss how it works and what it's best for.

John Angelo Yap
Updated December 13, 2023

a robot creating a new world, generated with Midjourney
Reading Time: 8 minutes
When it comes to generative AI, it’s hard to overstate the impact it has on society. Everything you need with just a single prompt — that’s revolutionary technology. But after a couple months, I have to admit that the novelty is wearing off a little bit. A huge part of that is because of the amount of AI tools we have today, which begs the question:
Do we really need another?
For AI image generators, we already have Midjourney, DALL-E, Firefly, and Stable Diffusion. Now, Meta’s entering the ring with their new image generator. This time though, something caught my eye: this application is completely free. So, of course, I had to try it.
In this article, I’ll discuss everything you need to know about Meta’s new AI image generator, the science behind its model, its output quality, and what purpose it serves in an already crowded space.
What Is It?
Meta’s AI image generator is a generative model that transforms prompts into images. It uses a model called Emu, which has scored 82.9% on visual appeal and 36.7% on text faithfulness.

However, it does come at a price. Emu is trained on over 1.1 billion images, all from Facebook and Instagram. Yes, you read that right — Meta uses photos uploaded by users to social media sites as their training set. It’s like what they say: If you’re not paying for the product, then you are the product.
Who Is It For?
Like other AI image generators, Meta is best used as a playground for creative hobbyists. If you’re an entrepreneur, you can also use Meta as a means to earn money passively. Bloggers and web developers can also use it to generate web content (featured images, stock images, vector art) quickly.
How It Works
Meta is much more barebones compared to Midjourney, DALL-E, and Firefly. It doesn’t have any features other than image generation. No aspect ratio control, no negative prompts, nothing. This makes for a more accessible and straightforward interface, wherein you just need to input a prompt, press the “Generate” button, and wait for a few seconds.

For prompts, you separate elements with a comma. The first parts must always contain the subject of the prompt (what you want to see) and the subsequent parts should have the context (what’s the look you’re going for). Each prompt will have four image variations. Simply choose what you like most and download them. The rest is up to you.
Meta’s Output Quality
Portraits
Prompt A: a man after a marathon, sunrise hour, closeup


Prompt B: a student in a class, golden hour, candid


I’m blown away at how good these are — to the point that I’m ready to say Meta’s the best AI image generator out there for portraits. It does have some issues, like going over the top with the drops of sweat in Prompt A but overall, it’s eerily realistic. However, this is to be expected since it’s trained on user data.
Personal Score: 5 out of 5
Landscape
Prompt A: mount fuji


Prompt B: a serene lake, golden hour, pine trees


These are all still pretty good. Maybe this is a bit anecdotal, but I did have issues generating images of Mount Fuji since Meta kept giving me pictures with fireworks in the background. The ones that didn’t had a few nuance issues. Take the Mount Fuji image on the right for example. Notice how the clouds are completely in the wrong place? The generated lake images had no issues though: the scenery is beautiful, reflections make sense, and the supporting context are all met.
Personal Score: 3.5 out of 5
3D Product Mockups
Prompt A: product photography of scented candles, sandalwood, earthcore, warmth, 8k, hd, ultra-detailed


Prompt B: product photography, a perfume, studio lighting, shadow play, lavender, soft


These prompts are the same ones I used in a recent article. Considering that I used Midjourney for that article, it’s remarkable how close the outputs are in quality especially since Meta is completely free. These are already market-ready product mockups, no need for further alterations.
Personal Score: 5 out of 5
Logos
Prompt A: simple vector logo of clothes


Prompt B: minimalist logo of furniture


I lost count of how many times I pressed the “Generate” button just to get good logos from Meta. The results speak for themselves: Meta just isn’t there yet. Out of these four, only the first clothing logo looks okay. The rest are barely recognizable and incoherent.
Personal Score: 1 out of 5
Abstract Concepts
Prompt A: failure


Prompt B: consciousness


I like feeding AI image generators with abstract concepts to test its nuance. Meta passed this test, in my opinion. The ideas are there, and you could infer how it’s trying to show these concepts with just a glance. However, I don’t think the images are well-executed. It’s trying to do too much, which results in rendering errors if you look closely.
Personal Score: 4.5 out of 5
High Context Prompts
Prompt A: a dramatic and cinematic illustration of a futuristic library engulfed in flames, giant robots, towering over the burning bookshelves, a clash with a group of librarians armed with archaic weapons and ancient knowledge, air is thick with smoke and dust as sparks fly and pages flutter in the inferno, a desperate struggle for the preservation of knowledge and the fight against a dystopian future, sci-fi


Prompt B: a vibrant and chaotic scene depicting a party amidst the apocalypse, the world around them crumbles, people gather in celebration, music blasting and lights flashing, desperation, joy, the bittersweetness of embracing the final moments as a black hole swallows the world


If we’re talking purely about nuance, Meta is almost perfect. It only skipped one or two elements within the prompt, which is a much higher success rate than Midjourney, but not as good as DALL-E. The images themselves are pretty good too, considering that I tried to cram as much information into the prompt.
Personal Score: 4.5 out of 5
Average Score
With all scores tallied up, my personal rating of Meta’s output quality is 3.92 out of 5. That’s a solid B+ in my book.
It’s also worth noting that Meta has an incredibly fast generation speed. Way faster than Midjourney, DALL-E, Stable Diffusion, and Adobe Firefly. This is remarkable considering that it’s going toe-to-toe with these established AI image generators already in quality. And again, this is completely free.
Pros And Cons
PROS | CONS |
|
|
Meta’s AI Image Generator Alternatives
Midjourney
Whenever someone mentions AI art, my mind immediately goes to Midjourney. It’s the most popular AI image generator today with more than 16 million registered users and counting. Its primary goal is to turn your ideas into beautiful pieces of art. Unfortunately, it still doesn’t have a platform, and you can only access it using their Discord bot.
Output Comparisons
For these comparisons, the ones on the left will always be Meta and the right will be Midjourney.


Portraits


Landscape


3D Product Mockups


Abstract Concepts


Logo
What Makes Midjourney Better Than Meta
- Way better at creating stylized artwork.
- Can use artist names and previous artwork as prompts.
- Comes with features such as an image upscaler, region variations, zooming, and panning.
- Allows you to generate images in different aspect ratios.
- Less prone to hallucination and rendering issues.
- Large community with continuous developer support and updates.
- It saves your previous outputs.
What Makes Midjourney Worse Than Meta
- Can’t create realistic images.
- Slower generation time.
- You can only access Midjourney through Discord.
- It’s paid.
DALL-E 3
DALL-E 3 is the current model of OpenAI’s image generator. It’s designed to be more creative and nuanced than its previous iterations. You can use it with ChatGPT, which processes your natural language prompts into ones that could be easily understood by DALL-E using GPT-4.
Output Comparisons
For these comparisons, the pictures on the left will always be Meta and the right will be DALL-E 3.


Portraits


Landscape


3D Product Mockups


Abstract Concepts


Logo
What Makes DALL-E Better Than Meta
- Incredible at nuance.
- Unmatched text generation capabilities.
- More customizable than Meta.
- Can be used with ChatGPT Plus for conversational prompts.
- It saves your previous outputs.
What Makes DALL-E Worse Than Meta
- Slower generation time.
- Worse at creating realistic images.
The Bottom Line
Don’t sleep on Meta.
They may be late to the game, but the value they provide considering that they’re a free image generator is immense. It even outperforms Midjourney and DALL-E when it comes to portraits — it’s insane how good this is.
That said, it’s plagued by the same issues that Meta as a whole has been dealing with for a while now: privacy. The reason why it’s better for realistic images in the first place is because they trained it using user images. So, maybe think twice before setting your images public on Instagram and Facebook from now on.
If you’re in the USA (sadly, it’s region-locked for now) and cool with everything above, then definitely give Meta a try.
Want to Learn Even More?
If you enjoyed this article, subscribe to our free newsletter where we share tips & tricks on how to use tech & AI to grow and optimize your business, career, and life.