Although DALL-E has made impressive progress since its debut, it still has significant shortcomings. Frequent users might notice several issues that require attention, including inconsistencies and sizing errors.
1
Still Can’t Get Text Right
While AI struggles with rendering text in images overall, I have found DALL-E to be particularly lacking in this area compared to other AI art generators I’ve tested. Simpler requests tend to yield better results, yet, as illustrated below, DALL-E still has difficulty, such as misplacing the accent in the word “Café.”
In several tests, I’ve noticed that DALL-E has misspelled words despite giving clear commands. In other instances, the text is completely warped. It also tends not to generate legible text for signage and similar items.
2
Doesn’t Resize Images When Asked
While DALL-E understands the concept of creating images, it struggles with resizing them upon request. For instance, when I’ve asked it to resize images for various blog posts, it simply regenerates the original picture each time.
Fortunately, cropping and resizing images is a breeze with many free tools out there, like Canva. Still, it would be nice to do this directly within DALL-E without needing external software.
3
Fails at Creating Photorealistic Images
DALL-E doesn’t struggle much with depth of field, but it has a tough time producing photorealistic images. For instance, I first asked it to create an image of a golden retriever in the mountains without specifying the picture style.
While the image was decent, it still had an artificial look to it. I then requested a more photorealistic rendition, but the result was even less convincing and a bit unsettling.
Utilizing the Photorealistic Custom GPT can enhance the results, but it does require a premium subscription. For those seeking photorealistic images, I recommend exploring alternative AI art generators like Adobe Firefly.
4
The Software Doesn’t Listen When You Tell It Not to Do Something
To craft better prompts in various AI tools, including Midjourney and ChatGPT, specifying what you do not want can improve the outcome. DALL-E, however, doesn’t always adhere to this guidance.
Time and again, I have instructed DALL-E to omit certain elements from my images, but the software continues to include them. While regenerating the image sometimes resolves the issue, it would be preferable for the software to simply follow the directive without the need for redoing.
In comparison, I have experienced this issue less frequently when using Adobe Firefly.
5
It’s Inconsistencies in Certain Areas
Many AI art generators have their weaknesses; for example, when testing Photo AI, I found it particularly poor at depicting hands. DALL-E also exhibits inconsistencies across various aspects.
Although DALL-E excels at creating general images, it doesn’t always accurately depict people interacting with objects. Moreover, it struggles with generating believable computer screens, often rendering a puzzling assortment of shapes instead. In some cases, asking it to turn off the computer screen might yield better results, though it’s not a guaranteed fix.
6
Image Styles Are Not Consistent
DALL-E can generate a wide variety of image styles, from paintings to more three-dimensional depictions. However, this variation in output can sometimes lead to a lack of consistency.
When requesting an image from DALL-E, it’s essential to be specific about the desired style. Otherwise, you risk receiving unexpected results, which may not be ideal if you’re aiming for a cohesive look across multiple images.
7
Artifact Sizing
When aiming for accuracy, ensuring that the objects in your image are proportionate to reality is crucial. Unfortunately, DALL-E sometimes fails to achieve correct sizing. For instance, it may depict a person holding a phone that appears overly large.
I initially believed that asking for a wider-angle shot would fix the issue, but it often exacerbated it, resulting in an unrealistic portrayal of the person. While experimenting with different prompts might help rectify this problem for some, I have not discovered a consistent workaround yet.
8
DALL-E Struggles With Creating External Materials
I initially thought DALL-E would excel at producing custom designs for calendars, greeting cards, phone backgrounds, and more. However, I quickly found that it doesn’t perform well in this regard.
For instance, when I attempted to create designs for birthday cards, I ended up with images that featured actual birthday cards within them, which was not my intention.
If you want to utilize DALL-E for such designs, I recommend avoiding phrases like “birthday card” in your prompts. Instead, clearly outline what you’re looking for, so you can adjust the size afterward.
Generating just the image in DALL-E and adding text in another app could be a more effective strategy.
9
Low-Effort Images When Asking for Multiple Options
At times, I find myself unsure about the specific AI design I want to use. In such cases, I often look to generators to create multiple variations of an image. However, in my experience, DALL-E tends to produce a lackluster array of options.
Frequently, the images I receive share striking similarities, which means I must be very detailed in my requests. I have also found that generating images one at a time yields better results than asking for a variety at once.
Overall, while I enjoy using DALL-E, it’s important to recognize that it isn’t perfect and that you’ll likely encounter some challenges. For instance, I’ve found it struggles to create photorealistic images unless using Custom GPTs, and it often fails to exclude unwanted elements from the generated images. Still, I encourage you to try DALL-E for yourself and see whether these issues arise in your experience.