Freepik's New ‘Mystic’ AI Image Model Is A Threat To Midjourney

For many years, Midjourney has held the crown as the best AI image generator on the internet. It has produced some of the most viral and controversial AI-generated images we’ve seen, like Pope Francis wearing a puffer jacket or Donald Trump getting arrested—memes that captivated and divided the online world.

Midjourney’s unmatched level of prompt coherency and photorealism has put it ahead of even the biggest names in AI, like OpenAI, Google, and Amazon.

Well… today, Midjourney just found a real competitor. The popular web-based design platform Freepik has announced its latest and most powerful AI image generator model yet — Mystic.

In this article, I’ll break down how Freepik’s Mystic stacks up against Midjourney’s V6.1. We’ll compare them based on three crucial criteria:

• Image quality
• Prompt coherency
• Text rendering accuracy

Let’s get started.

What is Mystic?

Mystic is Freepik’s latest and most capable image model—much more powerful than the latest Flux image models. When you open the image generator tool and expand the Model option, you’ll see ‘Mystic’ at the bottom of the list.

Right now, Mystic is still in the alpha phase and only accessible to a select few users. I was lucky to have early access to this model, and I am super excited to share my thoughts and experience with it.

Let’s start by comparing their abilities in one of the most important areas: prompt coherency.

Prompt Coherency

One of the most critical aspects of AI image generation, and often where many models fall short, is prompt coherency. This refers to the ability of an AI image generator to create images that are logically and visually consistent with the input prompt. Achieving this requires prompts that are clear, specific, and free of ambiguity or contradictions, ensuring that the generated image matches the intended output.

To evaluate the coherency of both the Mystic and Midjourney models, we used the following prompt:

Prompt: Photo of a red sphere on top of a blue cube. Behind them is a green triangle, a dog on the right, a cat on the left

Looking at the images above, Freepik’s Mystic performed impressively, accurately rendering all the elements as described and in the correct order. In contrast, Midjourney V6.1 struggled—not only did it fail to represent all the elements correctly, but it also placed the cat on the right side, contrary to the prompt.

While some may argue that Midjourney occasionally struggles with specific tests like this, my repeated trials showed consistent failures in this scenario.

This prompt was first introduced to showcase the coherency of Stability AI’s Stable Diffusion 3 image model. It’s quite surprising that a five-month-old SD3 image model still performs better than the latest model from Midjourney.

Here’s an example result from Stable Diffusion 3.

The accuracy of the result to the prompt is on point and the added subtle green tint on the white fur of the animals is very impressive.

Image Quality

Next, let’s talk about image quality. To really compare Mystic and Midjourney, I generated images using the following prompt:

Prompt: Cinematic photography, medium shot, glamour instagram young woman. She has brown hair with purple highlights. glamour woman. She has a pretty face, She is sitting in a coffee in bar. An influencer, smiling face, with hand gestures as if she’s explaining something

Both images look great! Mystic delivered a stunning image, with a focus on detail. The skin texture appeared more natural, and the fine strands of hair were more distinct, giving the image a lifelike quality that’s hard to beat.

Midjourney’s image was equally impressive at first glance. However, one recurring issue with Midjourney is its tendency to struggle with rendering fingers correctly. If you didn’t notice it, look again.

Let’s try another example:

Prompt: The baby kangaroo is unbelievably tiny, even smaller than a single grain of sand, perched delicately on the fingertip. Its minuscule body is covered in soft, velvety light brown fur, making it look like the tiniest, fluffiest speck of earth. The kangaroo’s eyes, though incredibly small, are bright and wide with curiosity, twinkling with a gentle innocence that melts the heart. Its tiny face, no bigger than a poppy seed, radiates pure sweetness, with an adorably round snout and a little smile that adds to its cuteness. The kangaroo’s tiny, rounded ears are perked up and impossibly cute, giving it an extra touch of charm.

Okay, both looked good here. When it comes to animal photos, Midjourney and Mystic tend to work incredibly well. However, the level of detail and realism that Mystic brings to the table is just a bit sharper, making the image feel more grounded in reality. Midjourney often leans toward more saturated and highly stylized results.

Text Rendering Accuracy

One important aspect of AI image models is their ability to render legible texts on images. This is often where image models struggle because, to AI models, text is just a combination of lines and shapes, not something with inherent meaning. They don’t have the same intuitive understanding of letters, words, and their arrangements that humans do. The AI sees text as just another visual element, not as symbols representing language.

Surprisingly, both Mystic and Midjourney are able to understand various surface textures and put text on them. Check out these results:

Prompt: An image of a young male model wearing a gray shirt with the text “Freepik Mystic” on it. The image is for advertising purposes with a plain background and studio lighting.

In this case, I would opt for the image generated by Mystic. Not only was the text more legible, but the overall composition of the image felt more professional.

Freepik’s Mystic VS Midjourney V6.1

Based on these comparisons, it’s clear that Mystic has a lot going for it. While we still don’t know which Flux model Mystic is based on or the specific datasets used to fine-tune it, the performance speaks for itself. Mystic consistently outshines Midjourney in areas where the latter has struggled — particularly in rendering limbs and fingers correctly, as well as placing text accurately on various surfaces.

Additionally, Freepik’s acquisition of Magnific.ai further enhances Mystic’s value. With the integration of the most powerful AI image upscaler, users can now upscale their AI-generated images with just one click.

This added convenience, paired with Mystic’s superior performance, makes it a formidable challenger in the AI image generation arena.

Final Thoughts

In conclusion, I have to say that Mystic feels like the better choice over Midjourney. The fact that it’s still not publicly available yet means that there could be further improvements in the pipeline before the final release.

It was surprising to see Midjourney finally opening up its web platform to everyone, signaling a shift away from its Discord roots—a move that, let’s be honest, a lot of non-tech people will appreciate.

But why did it take them this long, and why now? My guess is that they’re starting to feel the heat from competitors, especially with the release of Flux by Blackforest Labs. Platforms like Freepik have already integrated Flux, and it’s working seamlessly with other tools like Upscaler.

To stay at the top, Midjourney needs to release a more powerful image model soon. If they don’t, I wouldn’t be surprised if we start hearing a lot more about competitors like Freepik or Ideogram in the near future.

‍