Here’s why AI can’t type text into images—and how to work around it

If you’ve used generative AI tools to create images, you’ve probably encountered problems writing text on images. Knowing why this happens will go a long way to avoiding it, as well as knowing workarounds.

Why can’t artificial intelligence write text in images?

The biggest reason AI can’t write text is historical input. While AI is generally quite good at generating images, some aspects – such as text – have not been used to the same level for training LLMs (large language models). Because of this, AI is not that good at generating text in images.

Artificial intelligence tools are already quite advanced, but technically they are still in their infancy. Additionally, some AI tools run out of training data. It’s still possible that generative tools can be better at generating text, but for now you’ll have to try the workaround.

Are there workarounds for unreadable text in AI images?

Although generating text in AI images is quite challenging, it is not impossible. Here’s what I’ve found works for me…

1. Give the generator more specific challenges

When I first started using AI, my prompting skills were substandard at best. I’ve made the same mistakes when invoking ChatGPT that many people do, and I’ve also done it with other generative AI tools. My biggest mistake was giving vague prompts like “please create a picture of a street scene”.

Not surprisingly, I got substandard results. I think the best way to use generative AI is to think of your challenge as a creative brief. The more specific you are, the better. Here’s one challenge I tried, for example:

Create an image of the exterior of an Italian cafe. I want it to have “Café” as a sign and it should be on a sunny day.

As you can see from the image below, ChatGPT did a pretty good job. I don’t think it would be if I said “Create an image of the exterior of an Italian cafe”.

I also suggest creating simpler images; from personal experience these seem to work better. For example, the image above has fewer elements than what I tried below – meaning the AI ​​has less to try to figure out.

2. Use variants of the word “Text”

I have often found that a slight change in the language in my prompts can make a huge difference. After trying the term “text” a few times, it started to feel like I was going around in circles, so I tried trying different synonyms.

For example, instead of “text” you can try:

• Title
• Letters
• Written words
• Sign

If none of these work, try alternatives. What does and does not bring success can vary depending on what you are trying to create. For example, you probably wouldn’t use “sign” if you were trying to design a birthday card. The key here is to experiment with different AI ideas, and different generators can also work in different ways.

3. Add text using alternative tools

I wouldn’t recommend adding text using alternative tools if you want the words to be part of your image, but it’s a good idea if you’re trying to create more graphics. It will work if you want to design a card, graphic, book cover or something along those lines.

If you do this, I recommend making sure there is enough space on the image. You can use any tool you want to add text to images. I personally recommend Canva or Adobe Photoshop Express, but other apps that let you add text to smartphone images will also work for content you create with AI.

You should also make sure that the text matches well with your AI generated image. Use fonts that match the mood you’re trying to convey, and transform everything in any way you feel is necessary.

4. Keep the text brief

Every time I’ve tried to generate text in AI images (and failed) it’s often when I’m trying to add too much text. To be honest, this happens every time I try to create something with more than 10 characters. Just as I suggest keeping your images simple, I think you should do your best to keep your text concise.

To test this, I tried another challenge:

Please generate an image of a bank with the text “bank” on the outside. I want it to be downtown and a modern structure like what you would see in an American city.

As you can see below, the AI ​​tool did a pretty good job of listening. However, I noticed some flaws in one part of the picture. So I suggest telling the AI ​​that you only want one or two signs to limit the likelihood of this happening. These issues also seem to happen to me when the text is smaller, so you might want to keep that in mind.

5. Use the word correction tool

Just as you can remove blemishes on photos taken with a camera, you can try using a tool to help you fix words on images. There are plenty of apps that fix nonsense text in AI-generated images, such as Storia Lab.

Using these tools you can normally select text and change what it says accordingly. Some apps are free, but in other cases you will need a paid subscription. If you use AI to generate a large number of images, it may be worth purchasing a monthly or annual access plan whenever you need it.

6. Use an artificial intelligence generator that produces accurate text

You’ve probably heard of Midjourney, DALL-E, Firefly, and similar AI image generators. But did you know that there are specific AI art generators that create accurate text in images?

Since one of Ideogram’s specialties is generating readable text, it makes perfect sense to at least try out the application for a few of your creations. The ideogram has a A magical challenge a feature that extends your original one, meaning you can get even more accurate results. In addition to the free plan, paid subscriptions start at $8 per month.

Despite the shortcomings of generative AI when creating images with readable text, you’ll find a few solutions if you’re willing to think outside the box. Sometimes the best option is to give your tool more detailed prompts but fewer words to solve. You may also want to fix image problems with other software or use tools specifically designed to create accurate text.