I’ve spent a lot of time living inside the world of AI image generators, and let me tell you, it’s a fever dream. One minute I’m watching a stunning, cinematic sci-fi scene that looks like it cost millions to make, and the next, I’m watching a “funny” disaster where someone has three arms and teeth where their eyes should be. Tools like Midjourney, DALL-E, and Stable Diffusion are developing at breakneck speed, but they still have some unique blind spots.

Don’t get it wrong—I use these tools every day to create everything from realistic product shots to stylish avatars of myself. The technology is revolutionary, but after thousands of prompts, you start to see cracks in the code. Despite all the “pro” updates and massive data sets, there’s a certain list of things the AI ​​still fails with embarrassing consistency.

If you’ve ever tried to create a simple group of people or a specific piece of text and ended up with a horrible mess, you’re not alone. Even the most powerful planetary models still struggle with basic physics and anatomy at first. I know the recurring traps that disrupt even the smartest algorithms, so you can stop wasting your credits on prompts that are doomed from the start.

Don’t miss any of our unbiased tech content and lab-based reviews. Add CNET as a preferred source of Google.

Some AI image programs give you tools to edit out these errors, which is great. I recommend taking advantage of them whenever possible. For programs that don’t offer tools, or more likely, if tools don’t solve the problem, this is what you should do next.

Based on my experience, I’ve compiled my tried-and-true tips for tweaking your prompts and settings to optimize your images. For more, see best AI image generators and our guide to effective AI image prompt writing.

Human faces and expressions

Accurate facial expressions continue to challenge AI generators. Unique eyes, teeth and eyebrows are some of the strongest indicators that an image is AI-generated. In this case, the result is too funny for me, if not too useful. The girls are sporting some Halloween-like vampire teeth, and the dude in the back is having more than one bad hair day.

Even with cartoon or unrealistic characters, generators struggle with moderate emotions and expressions. This image — produced by our top pick, Dall-E 3 — is over-amplified at the prompt, and the result is extremely dramatic. I’m a self-proclaimed neat freak, but I can’t imagine anyone getting mad at what hundreds of dollars worth of cleaning supplies look like. Even the best programs can fail and produce bad results.

How to fix it: I recommend asking the service to reduce the number of people it tries to render – thus reducing the chances of error – and using post-generation editing tools to select specific parts of the image that need to be changed or fixed. Choosing a milder adjective (“angry” instead of “angry”) may help steer the service in the right direction.

Iconic characters, trademarks and recognizable logos

For many reasons, AI image generators rarely produce accurate logos, well-known trademarks or otherwise unrecognizable characters. This is often because there are significant legal concerns about a company allowing its brand materials or intellectual property to be used for AI services. It can also be because the logo or brand you want to include is not in the generator’s training data, so the generator doesn’t understand your request.

Two recent exceptions to this are the new image creation features of Google Pixel 9 and Pixel 9 XLthanks to Gemini AI, used by our mobile experts make pictures featuring accurate-looking renderings of Mickey Mouse and Pikachu. The other exception is some paying X (formerly Twitter) users who report that they can use the Grok chatbot on the social media site to make realistic looking Mickeys and political figures. These images are still not perfect renderings, but they are close.

How to fix it: You can’t, but on the plus side, you’re more likely to avoid copyright infringement lawsuits. If you have this issue, I recommend revisiting your design concept to see if you can get the same point across without using a specific company name or logo. Do you really need the TikTok logo, or do you need a phone with a vertical video player, for example.

Overlapping, complex elements

Sometimes, you can stump a generator if you have too many overlapping or complex elements within an image. The best AI image programs don’t always run into this issue, but even the best programs fail at some point. With this image of Leonardo, the library of my dreams was destroyed by a rolling staircase that disappeared in the middle.

Similarly, this photorealistic image of a kitchen looks good from a short glance, but a second look or a quick zoom-in shows the nonsensical characters printed in the cookbook, and the book itself seems to have two spines and three sections. Images like this show how frustrating AI image generation can be — even when you think you’ve got a winner, small mistakes make it unusable.

How to fix it: Try simplifying your prompt or using post-generation editing tools to select the problem and fix it. Sometimes changing the aesthetic helps too; these issues usually arise when you are trying to create photorealistic or stock images. If your service offers a specific area editing tool, try using that to select the questionable area and ask the generator to remove what’s in it.

Excessive editing and hallucinations

The hallmark of a good AI image generator is its ability to create an image with few errors or assumptions while giving you the right tools to solve issues when they inevitably arise, but not all services. In this example from Midjourney, I went through several rounds of edits on this image of a soccer team celebrating a victory, and I don’t know what happened to get this final result – or any indication of what happened to the blob that is probably a player on the right side of the photo. Honestly, I don’t think Midjourney knows either.

How to fix it: With AI image editing, sometimes less is more. Don’t be afraid to scrap your current batch of images and start over. You can always preemptively fix big issues by refining your prompt so you can only worry about fixing small issues later.

AI generators still require human work

I have no doubt that the companies behind these AI generators are working to eliminate these issues and level up their models. For now, the imperfect images show us again that while this technology is improving, it is not perfect. My hope is that since many of these services are facing similar challenges, they will all find solutions for these issues.

As a final note, I would remind you that it is a best practice to credit or acknowledge that an image was created by AI when you share it. As AI images get better and more realistic looking, it’s more important than ever to distinguish AI-generated art from other types of media.

For more, see our advice for writing the best AI image prompt and our best choice for AI image generators.