Expert-backed ways to fix AI image hallucinations when using ChatGPT, Midjourney and more

As CNET’s reviewer for AI image and video generators, one of the biggest perks of my job is laughing at the hilarious, extremely bizarre, and sometimes frightening images that these generator tools spit out.

Make no mistake, AI creation software has come a long way in a short time. I’ve created stunning sci-fi scenes in Midjourney, realistic product images with Stable Diffusion, and powered up AI with ChatGPT. But no AI service is perfect, and I observed some notable trends during my testing.

There are some things that, despite constant updates, AI image generators just can’t handle well the first time.

Some AI image programs give you tools to edit these errors, which is great. I highly recommend taking advantage of it as much as possible. For programs that don’t offer these tools, or more likely, when these tools don’t fix the problem, here’s what you should do next.

Based on my experience, I’ve compiled my proven tips for tweaking your prompts and settings to optimize your images. To learn more, check out the best AI image generators and our guide to effective AI image prompt writing.

Human faces and expressions

Accurate facial expressions continually challenge AI generators. Weird eyes, teeth, and eyebrows are among the strongest indicators that an image is AI-generated. In this case, the result seemed extremely funny to me, if not completely unusable. The girls are sporting Halloween-worthy vampire teeth, and the guy in the back is having more than just a bad hair day.

Even with cartoon or non-realistic characters, generators have difficulty moderating emotions and expressions. This image – created by our top pick, Dall-E 3 – over-amplified the prompt and the end result was overly dramatic. I consider myself a neat freak, but I can’t imagine anyone being that upset over what feels like hundreds of dollars’ worth of cleaning supplies. Even the best programs can fail and produce shaky results.

How to fix it: I recommend asking the service to reduce the number of people it tries to render – thereby reducing the chance of error – and using post-generation editing tools to select specific parts of the image that need to be regenerated or corrected. Choosing a gentler adjective (“angry” rather than “enraged”) could help guide the service on the right path.

Iconic characters, recognizable brands and logos

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

Two recent exceptions to this rule are the new image generation features on the Google Pixel 9 and Pixel 9 XL, thanks to Gemini AI, which our mobile experts used to create images featuring accurate renderings of Mickey Mouse and Pikachu. The other exception is some paid users of X (formerly Twitter) who say they can use the social media site’s Grok chatbot to create realistic human Mickeys and political figures. These images are still not perfect renderings, but they are pretty close.

How to fix it: You can’t, but on the plus side, you have a much better chance of avoiding copyright infringement lawsuits. If you’re having this problem, I recommend reviewing your design concept to see if you can get the same message across without using a specific company name or logo. Do you really need the TikTok logo, or do you just need a phone with vertical video playback, for example.

Complex and layered elements

Sometimes you will crash a generator if you have too many overlapping or complex elements in an image. The best AI imaging programs won’t encounter this problem very often, but even the best programs fail at some point. With this image by Leonardo, the library of my dreams is marred by the rolling ladder that disappears halfway up.

Likewise, this photorealistic image of a kitchen looks okay at a glance, but a second glance or quick zoom shows absurd characters printed on the recipe book, and the book itself appears to have two spines and three sections. Images like this show how frustrating it can be to create AI images: even when you think you have a winner, small flaws render them unusable.

How to fix it: Try simplifying your prompt or using post-build editing tools to identify the problem and fix it. Sometimes changing the aesthetic can help too; These problems usually appear when trying to create photorealistic or stock images. If your service offers a specific zone editing tool, try using it to select the questionable zone and ask the generator to remove what is in it.

Overediting and hallucinations

The sign of a great AI image generator is its ability to produce an image with few errors or hallucinations while giving you the right tools to fix these problems when they inevitably appear, but not all services do this. In this example from Midjourney, I’ve made several rounds of edits on this image of a football team celebrating a victory, and I have no idea what happened to get to this end result – nor any idea what’s going on with the blob that is presumably a player on the right side of the image. Frankly, I don’t think Midjourney knew either.

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

AI generators still need human work

I have no doubt that the companies behind these AI generators are working to eliminate these problems and improve their models. For now, imperfect images show us once again that while this technology is improving, it is not flawless. I hope that because many of these departments face similar challenges, they will all look for solutions to these problems.

Finally, I remind you that it is recommended to credit or acknowledge that an image is generated by AI when you share it. As AI images improve and become more realistic, it is more important than ever to distinguish AI-generated art from other types of media.

To learn more, check out our tips for writing the best AI image prompt and our top picks for AI image generators.