DALL-E, OpenAI’s powerful AI image generator, has captured the imagination of creators, designers, and curious users around the world. It’s capable of producing astonishing artwork in seconds, interpreting natural language prompts to create visuals that would’ve once required hours of human effort. But despite all its innovation and creative potential, DALL-E is far from perfect.

The more one uses the platform, the more apparent its limitations become. From generating inconsistent imagery to struggling with basic instructions, DALL-E still has a long way to go before it can be considered a reliable tool for all use cases. Here’s a closer look at the 9 biggest problems with DALL-E, based on extensive experience and experimentation.

1. DALL-E Still Struggles With Text in Images

One of the most annoying things about DALL-E is that it can't properly handle text inside pictures. A lot of AI drawing tools have this problem, but DALL-E usually does worse than others in this area.

Users frequently encounter misspelled or distorted text, even when the prompt is extremely specific. A common example is the word “Café,” which DALL-E often misrepresents, dropping accents or generating unreadable typography. Similarly, attempts to create street signs, posters, or logos with legible writing often result in gibberish or warped characters.

2. It Can’t Resize Images as Instructed

Another glaring flaw is DALL-E’s inability to resize images properly when asked. Users looking to generate visuals in specific dimensions—say for blog banners, social posts, or posters—often find that DALL-E doesn’t actually resize the image but rather regenerates a brand-new one without considering the requested aspect ratio or format.

While image resizing can be done through third-party tools like Canva or Photoshop, it would be far more convenient if DALL-E followed those instructions. For a tool that’s praised for its versatility, this limitation feels like a step backwards.

3. Lack of Photorealism in Output

Despite improvements in image realism, DALL-E often struggles to generate convincing photorealistic images. Even when prompts explicitly request natural or life-like visuals, the results frequently veer into the uncanny valley.

For instance, an image of a golden retriever in the mountains might look fine at first glance, but the more one inspects it, the more obvious it becomes that the image is AI-generated. Attempts to increase realism often backfire, making the image look even less believable and sometimes even unsettling. Users who require true-to-life visuals often find better success with tools like Adobe Firefly or Midjourney, which tend to perform better in photorealistic rendering.

4. Ignores Instructions on What Not to Include

In prompt engineering, a common best practice is to specify what you don’t want in the image—for example, “Do not include people in the background” or “Avoid showing text.” Most modern AI tools understand these instructions to a reasonable extent.

DALL-E, however, seems to struggle significantly with negation. Even after explicitly telling it to exclude certain elements, users often find that those very things end up in the final image anyway. While regenerating can sometimes solve the issue, it adds extra time and frustration.

5. Inconsistencies in Object Representation

Like many AI art generators, DALL-E has a few weak spots when it comes to specific subjects. In particular, it shows inconsistencies when generating hands, faces, and certain objects like computer screens or gadgets.

In one case, when prompted to create an image of a person holding a smartphone, DALL-E struggled to scale the phone properly, often rendering it comically oversized. Similarly, computer monitors often appear as jumbled rectangles with nonsensical details.

6. Style Inconsistency Across Generated Images

One of DALL-E’s most versatile features is its ability to create artwork in various styles—cartoonish, painted, 3D, abstract, and more. However, this same versatility becomes a weakness when the tool fails to maintain a consistent style, especially across multiple outputs within a single project.

For example, users aiming for a cohesive visual theme across different slides, graphics, or blog posts may find that DALL-E delivers images with wildly different artistic directions. Unless the prompt is painstakingly specific, the style can shift drastically between generations.

7. Poor Artifact Sizing and Perspective

Another common problem with DALL-E is the incorrect sizing of objects in relation to their environment. When prompted to create scenes involving people interacting with objects, the AI often gets artifact proportions wrong.

Phones appear too large, books may be outsized, and drinks can look like gallon jugs in someone’s hand. Attempting to “fix” this by adjusting the angle or requesting a wide-angle shot sometimes makes the image look even more unnatural, distorting human figures or background elements in the process.

8. Difficulty Designing for External Use

Many users turn to DALL-E for personalized design ideas—custom birthday cards, calendars, wallpaper designs, or social media graphics. Unfortunately, DALL-E often misinterprets the intent of such prompts.

For instance, when asked to create a birthday card design, the AI may generate an image of an actual greeting card rather than a design intended to be printed on one. It creates confusion and misalignment with the user’s goal.

9. Low Effort in Multi-Option Generations

When users ask DALL-E to create multiple versions of an image, the output often feels uninspired. Instead of receiving distinct creative options, users tend to get minor variations of the same image. While this might be helpful in refining a concept, it doesn’t offer the diversity that most would expect when requesting several options.

Many find that generating images one at a time and tweaking each prompt slightly leads to better results. Still, the fact that batch generation feels low effort defeats one of DALL-E’s biggest selling points: speed and convenience.

Conclusion

DALL-E remains a remarkable innovation in the world of AI art, and its accessibility has helped countless users bring their ideas to life. But despite its popularity and power, it still suffers from significant limitations.

From struggling with text generation and ignoring user instructions to producing inconsistent styles and failing in photorealism, the tool has room for improvement. Many of these flaws become apparent only after extended use, which can be frustrating for professionals relying on precision and polish.