Testing Midjourney: What Happens When You Ask for "Ugly" AI Art?

A Midjourney generated image using Midjourney Automation Suite

Have you ever wondered if you can tell Midjourney *not* to do something? As AI art tools get better, they often lean towards creating images that are visually appealing or technically proficient. But what if you want to go against that? What if you try to create something purposefully messy, blurry, or just plain ugly?

That was the idea behind a recent experiment diving into "anti-prompts" for Midjourney v7. The goal was to see how well Midjourney handles prompts that ask it to contradict its usual polished style or create results that break traditional rules of composition and aesthetics.

Midjourney often interprets prompts based on the words it recognizes, sometimes seemingly ignoring the negative or contradictory parts. Let's look at what happened when challenging the AI to make things "ugly" or "wrong."

Trying to Make Midjourney Art Look Bad

The experiment involved giving Midjourney prompts that included a main subject (like a forgotten civilization or a luxury spaceship interior) combined with an "anti-prompt" phrase (like "make everything ugly" or "blurry and poorly framed photo"). The prompts were often longer than usual, which itself can be a challenge for Midjourney, as it sometimes doesn't seem to process every single word.

Let's break down some of the attempts and results.

Results from Specific "Anti-Prompts"

Blurry and Poorly Framed Photo

Prompting for shots that look accidental, rushed, or off-center actually worked well. The results showed blurry images, off-kilter framing, and visual noise that gave a sense of imperfection. Despite the different subjects, there was a consistent feel of motion and disorientation. Midjourney seemed to understand this request effectively.

Ugly Composition, No Rule of Thirds

This prompt didn't quite hit the mark. Instead of avoiding the rule of thirds, many images actually *used* it. The images weren't exactly beautiful, but they weren't truly ugly either – more just ordinary. Midjourney seemed to focus on "composition" rather than "ugly composition" or "no rule of thirds."

Overexposed with Washed Out Colors

These results matched the prompt nicely. The images looked overly bright with highlights blown out and faded colors. There was a consistent softness and lack of contrast that successfully created the requested look. Most images across different subjects adopted this washed-out style.

Underexposed, Barely Visible

This attempt mostly missed the target. While the lighting was low and shadows were present, details were still clear. The images looked more like moody, stylized shots than genuinely hard-to-see underexposed photos. Midjourney went for cinematic mood rather than true invisibility.

Flat Lighting, No Contrast

Most of these images failed to show flat lighting. There was still noticeable contrast, directional lighting, and distinct shadows. Even in attempts at even lighting, the results looked more cinematic than truly flat or lifeless. Midjourney's default tends towards dynamic lighting.

Random Lighting Sources, No Shadows

Again, Midjourney mostly showed controlled, realistic lighting with shadows. This contradicted the prompt for random or shadowless scenes. The AI seemed to prioritize creating coherent, cinematic lighting setups instead of unpredictable ones.

Messy Background with No Focal Point

This worked better. Many scenes felt cluttered and visually noisy without a clear subject. However, some images, especially in the fashion and interior styles, still had strong subjects and structured layouts, showing mixed results depending on the core subject.

Unbalanced Visual Weight

Midjourney struggled with this concept, similar to the "no rule of thirds." Most images showed balanced compositions. The AI appears to have difficulty picturing scenes that are intentionally unbalanced or unstable.

Too Much Negative Space in All the Wrong Places

This had mixed success. Some images, particularly landscapes, effectively showed subjects lost in vast empty spaces with awkward spacing. Others, like portraits, remained traditionally composed and centered. So, it worked sometimes but not consistently.

Everything is Slightly Off-Center

Despite the prompt, Midjourney consistently produced images that were strongly centered or symmetrically composed. The framing felt deliberate and polished, missing the intended subtle imbalance.

Unusable User Interface for a Futuristic App

The first attempts looked sleek and modern, not unusable. Better attempts later on showed cluttered, overly complex layouts prioritizing style over function, which fit the "unusable" idea better. They looked impractical or confusing, even if still visually striking.

Terrible Product Design, No Ergonomics

Midjourney captured this concept well. The created objects looked stylish but clearly uncomfortable or impractical. Things like awkward chairs or oddly shaped items successfully showed form prioritized over usability.

Clothing That Doesn't Fit and Makes No Sense

Early attempts sometimes looked like avant-garde fashion rather than nonsensical clothing. Later attempts did lean into absurdity, showing wildly impractical, oversized, or jumbled outfits that genuinely fit the prompt better.

Architecture That is Completely Impractical

This worked well. Midjourney generated structures that looked absurd, unstable, or clearly not designed for living or safety. The resulting buildings were often visually striking but impractical, matching the prompt precisely.

Unappetizing Furniture That Hurts to Sit On

The images successfully created furniture that looked uncomfortable, made from harsh materials or awkward shapes. They appeared more like sculptural pieces to admire than sit on, fulfilling the concept of design over comfort.

Unreadable Book Cover with Chaotic Typography

Midjourney delivered visual noise effectively here. The covers featured overlapping, distorted text and textures that made reading difficult. They became more abstract statements than functional covers, well matching the chaotic and unreadable request.

Confusing Road Signs That Lead Nowhere

This concept worked quite well. The signs were often contradictory, excessive, or placed in surreal, desolate environments, creating a sense of confusion and lack of direction.

Nonsensical Control Panel for an Alien Ship

Many results looked more like complex retro-human tech than truly alien or nonsensical. Heavy on industrial switches and chunky dials. While some images achieved a strange vibe, a lot still felt vintage sci-fi rather than otherworldly confusing.

A Website from 2040 Designed in MS Paint

Midjourney captured the MS Paint aesthetic (bold outlines, bright colors, pixel art) but didn't make it look like a website. The results were more like stylized sci-fi postcards or comic panels – visually fun, but missing the actual web layout aspect.

Unappetizing Food Styling for a Gourmet Dish

Midjourney's tendency to beautify showed here. The dishes looked elegant and polished, even with unusual elements. They appeared more like creative fine dining or art projects than truly unappetizing food.

Cheerful Horror Movie Poster

This contradiction worked well. Midjourney blended creepy elements with bright colors and cartoonish styles, creating posters that were unsettlingly energetic and unique. The tension between cheerfulness and horror was captured effectively.

Romantic Moment with Hostile Body Language

Midjourney successfully captured the tension. Scenes showed romantic settings with figures whose expressions or posture indicated underlying conflict or guardedness, creating a feeling of a quiet standoff within warm lighting.

Child's Birthday Party with Noir Lighting

This worked beautifully. The noir lighting added drama and shadows, turning innocent birthday scenes into something cinematic and mysterious. It felt like a party caught in a dream or an investigation, adding an odd, gripping quality.

Tragic Funeral in Bright Pastel Colors

This combination created a surreal emotional contrast. Somber funeral scenes were rendered in soft, candy colors. The bright palette didn't hide the grief but highlighted it, making the sadness feel even more poignant in a dreamlike way.

Relaxing Spa Scene with Panic and Urgency

Most images missed this prompt. The spa settings were serene and relaxing, as requested, but they lacked the element of panic or urgency. Midjourney focused on the calm atmosphere and failed to introduce tension consistently.

Navigating complex prompts and ensuring Midjourney understands your exact intent, including contradictory ideas, can be challenging. Sometimes, breaking down your vision or using specific parameters helps. The Midjourney Automation Suite from TitanXT is designed to simplify this process, helping you manage prompts and variations more efficiently.

Happy People in a War Zone, Smiling

This worked compellingly. Midjourney showed calm or smiling expressions amidst devastation, suggesting resilience, relief, or moments after chaos. It didn't feel absurd but rather like people surviving or breathing again after hardship.

Crying Clowns in a Corporate Meeting Room

This hit the surreal satire vibe well. The clash of clown costumes with sterile corporate settings created a visual metaphor for workplace despair, often looking like dark comedy or surreal dystopia with performance art energy.

Wedding Ceremony with Nobody Looking at Each Other

Most images displayed typical weddings where people *were* looking at each other. Midjourney struggled with the concept of intentional disconnection in this common scene type.

Celebration with No Joy

Many images matched this, showing festive settings (balloons, costumes) but with emotionally flat or detached expressions. However, some group shots still showed clear happiness, so the results were mixed but often successful in conveying the contrast.

Serene Landscape with Screaming People

The serene landscapes were captured well, but the "screaming people" element was largely missing or too subtle to notice. Figures were present but usually depicted as calmly walking or standing, failing to convey panic or urgency.

Perspective Doesn't Make Sense

Most images presented plausible perspectives. While some had unusual placements or optical effects, few truly broke spatial logic enough to fully match the prompt for nonsensical perspective. Midjourney defaults to coherent spatial arrangements.

Impossible Physics, Floating Shadows

Impossible physics like floating objects were depicted quite effectively. However, the "floating shadows" aspect was less consistent. While lighting could be dramatic, shadows rarely appeared truly detached or behaved impossibly.

Faces That Keep Folding Into Themselves

This worked strongly. Midjourney created origami-like or layered faces with recursive folds, effectively capturing the idea of internal collapse or self-layering through visual transformation.

Objects Phasing Through Each Other

This concept was largely missed. Images were visually interesting with layered textures or surreal forms, but objects rarely appeared to be genuinely passing through one another as requested. The results lean more toward artistic abstraction than literal phasing.

Scene That Loops Onto Itself Recursively

This was a complex prompt. While many images featured repetition or looping shapes, few clearly depicted true recursive structure. The concept was suggested but not consistently or effectively illustrated through visual looping.

Reflections That Don't Match the Scene

Midjourney failed on this one almost completely. Reflections consistently matched their environments, showing realistic continuity. Even with slight surreal twists, they still echoed the primary subject too closely to be considered a mismatch.

Time of Day Constantly Changing

Images hinted at different times of day through varied lighting but did not show continuous transitions. Each frame was static, capturing moments rather than depicting change *over* time. The concept was implied visually but not expressed dynamically.

Gravity Only Works on Certain Items

This had mixed success. Some images showed floating items or people suggesting selective gravity. But many lacked clear signs of altered gravity, especially in structured interior scenes, making the overall result inconsistent.

Clouds Inside Buildings, No Windows

Midjourney depicted clouds indoors well. However, it consistently added windows or open facades to the buildings, directly contradicting the "no windows" part of the prompt. The surreal clouds were there, but the lack of windows was not.

Staircase That Ends in the Sky But Goes Nowhere

This prompt worked very well. Midjourney effectively generated staircases rising into the sky or clouds with no connection point, visually emphasizing the idea of leading nowhere in a surreal, dreamlike way.

Hyperreal Cartoon

This blending of styles was highly successful. Midjourney combined exaggerated cartoon features with realistic textures and lighting, creating a vivid, quirky, and slightly surreal look that fit the aesthetic perfectly.

Photorealistic Glitch Art

This hybrid style worked very well too. Realistic rendering was disrupted by digital artifacts like pixelation and data corruption, successfully creating a striking glitch aesthetic that contrasted realism with distortion.

Minimalism with Extreme Complexity

This worked well. Midjourney balanced sparse backgrounds with strikingly intricate central elements, creating tension between space and dense detail, effectively delivering both parts of the aesthetic.

Vibrant Monotone Colors

Midjourney captured this precisely. Images emphasized single dominant colors applied boldly while the rest of the scene remained muted, creating a striking and cohesive visual identity within one color family.

A Detailed Abstract

Some styles, like biomechanical or fever dream landscapes, aligned well with the request for high intricacy and rich textures in abstract forms. Others, like structured interiors, lacked the organic chaos needed, resulting in a partial match overall.

Chaotic Symmetry

This worked well. Midjourney combined mirrored or radial patterns with high detail and irregular forms, creating compositions that had underlying structure but incorporated unpredictable, complex elements.

Perfectly Disorganized Still Life

This concept was well represented. Scenes captured cluttered arrangements that still felt intentional and aesthetically composed, balancing scattered objects with good lighting and detail. It felt like controlled chaos rendered with precision.

Completely Random Design with Calculated Intention

This paradoxical concept worked impressively. Midjourney created abstract or chaotic-looking scenes that still revealed underlying precision through symmetry, alignment, or balance, successfully blending perceived disorder with thoughtful arrangement.

Low-Quality Ultra HD Photo

This failed completely. Midjourney produced extremely sharp, polished, and high-quality images with no signs of the requested low quality, blur, or artifacts. The AI defaulted to impeccable rendering, ignoring the "low-quality" instruction.

Handcrafted AI Generated Item

This had mixed results. Some elements like sculptural forms or textiles evoked a handcrafted feel with physical textures. Others leaned towards clean digital renderings, lacking the intended tactile imperfection. It worked better when organic material detail was emphasized.

Key Learnings about Challenging Midjourney

The experiment confirmed that Midjourney doesn't process negative instructions like a human would. Telling it "no" or asking it *not* to do something often results in it focusing on the concepts mentioned and doing them anyway. It often struggles with instructions that ask it to break fundamental visual rules it seems built to follow, such as:

• Avoiding good composition (like the rule of thirds).

• Creating genuinely unbalanced or non-centered images.

• Producing truly flat or random lighting.

• Generating misaligned reflections or nonsensical perspectives.

• Making traditionally beautiful subjects (like food) look unappetizing.

• Introducing low quality or imperfections into its high-definition output.

Midjourney seems to have a built-in "prettifying" effect. Even when trying to make things intentionally ugly or messy, the results often come out looking stylized, cinematic, or visually interesting in some way. Achieving truly bad or non-aesthetic results might require more specific parameters like `--style` or `--weird` combined with the prompt.

Conclusion

Exploring "anti-prompts" in Midjourney is a fascinating way to understand the AI's strengths and limitations. While it excels at blending contrasting styles (like cheerful horror or noir birthdays) and creating visual paradoxes (like chaotic symmetry), it struggles with instructions that ask it to simply produce poor quality, poor composition, or illogical physical states consistently.

It's a reminder that prompting is not like giving instructions to a person. It requires understanding how the AI interprets words and concepts. For complex projects or when trying to achieve specific non-standard results, managing your prompts effectively is crucial. A tool like the Midjourney Automation Suite from TitanXT can streamline your workflow and help you experiment with variations and parameters more easily, making the prompting process smoother whether you're going for beautiful or intentionally bizarre.