ChatGPT’s Image Editing Feature – An Honest Lab51 Review

Last week, ChatGPT introduced its new feature that brings image editing to a completely new level. Now, users can generate and edit images directly within the ChatGPT interface. This functionality is available to users across various subscription tiers, including Free, Plus, Pro, and Team plans.

Couldn’t ChatGPT generate images before?

Yes, it could. However, generating and editing images was working with the integration of DALL·E 3 directly into the platform. Before this native integration, image generation was possible through plugins or external tools, but it wasn’t as seamless or directly embedded in the chat interface. 

With the newest update, ChatGPT (especially in its Pro version using GPT-4-turbo) gained built-in capabilities to create and edit images based on text prompts, including tasks like inpainting (editing specific parts of an image), making visual content generation much more accessible.

As the initial hype around the news begins to fade, the Lab51 Team decided to examine new features of ChatGPT, evaluate them, and give an honest review of the new image-editing features.

Remove or replace objects 

According to ChatGPT, the AI model can now seamlessly remove objects. The user just has to upload an image and write a prompt, such as  “remove the logo” or “swap the bottle for a rose”, and then ChatGPT generates an image according to the user’s request. Let’s test it!

We have uploaded a photo of the park to ChatGPT and asked it to remove people. A simple task, let’s see how the AI model will solve it.

The results are indeed outstanding! No people are in the picture while the original features of the image are kept.

Tip: ChatGPT generates square pictures by default. If you want to keep the original format, write it down in the prompt.

Let’s ask him a harder task. To replace an object.

We uploaded an image of a hand holding a cup of coffee and asked ChatGPT to replace it with a red apple.

The results were stunning!

However, on a closer look, the image details are slightly different from the original. That’s because ChatGPT generates the whole image from scratch. There is no feature (yet) that allows you to choose an area on the image and edit it, without generating the whole picture.

Tip: If you need to edit only a certain area on the picture without changing the whole image, you may consider using editors such as Photoshop Firefly, or Midjourney.

Change text 

According to the ChatGPT press release, they have improved the text-on-the-image feature. As we know, generating text on the generated image used to be a struggle. As a rule, AI models, including ChatGPT, were generating unreadable or nonsensical text. Now, this problem is supposedly resolved, and the text on the images must look decent. Let’s check it! 

We are bringing the task to the next level. The input picture contains a Spanish phrase and a Spanish flag. We asked ChatGPT to change the text to English translation and replace the flag with the UK one.

The results are amazing. Translation is correct and, moreover, the text font is very similar to the original.

Change facial expressions, makeup, and hair

That feature must be very useful for many fields such as photography, advertising, fashion, and the cosmetics industry. Let’s fully put it to the test and see what it can really do.

We asked ChatGPT to open the woman’s eyes, make her smile, and change her hair to ginger.

The results are not satisfying at all. As with many AI models now, editing a human’s face creates the uncanny valley effect, as it seems in the output picture. Many facial features were not kept, which is not good. And the image is not realistic, unlike the original. Indeed, human face generation still seems like a weak spot for many AI models.

Switch background 

This task seems to be challenging at first sight. According to ChatGPT, now the AI model can perform a relatively hard task by changing the background of the image, for instance, changing the studio to the beach, or plain to the luxury interior. Let’s try it out!

We asked ChatGPT to replace the background on the image from the streets to the garden while keeping the person. Let’s see the results:

The background was changed indeed. However here we encounter the same problem as with the previous task. ChatGPT AI model still has trouble with generating humans. The man in the output pictures has lost his features, looks uncanny and unnatural.

Tip: If you want to change the background, you can use ChatGPT first, and then use Photoshop or Canva to remove the background and keep the object at the foreground of the original picture, and then combine both photos.

Social Media Content Creation

ChatGPT claims that it can now produce content for social media, such as carousel visuals, memes, infographics, or content from scratch. Also, it can repurpose one image for multiple platforms (e.g., reformatting for IG Stories, FB posts, LinkedIn).

We asked the AI model to create a new design for a Home Page for the Lab51 website, describing what we want to see. Moreover, we uploaded our logo to see what it can do. 

The results were underwhelming. The output design felt unoriginal, basic, and dull. Nevertheless, there might be some potential ideas for inspiration from such a design.

As a bonus, we decided to try out the other feature of creating infographics. So far, we haven’t seen any AI image generating model that can decently create infographics. As a rule, AI generated an image that looks like infographics at first sight, but neither data nor visuals made sense at all. We asked ChatGPT to create infographics for Web3 financial investing in 2025.

The results were surprising. It is not flawless, it is simple, but it is much better than it used to be. ChatGPT collected information and made it into uncomplicated infographics. The text is clear, and the images more or less correspond to the context. Most probably, in the near future, AI-generated infographics will take over the market.

Styling image

We approached the most interesting feature that was hyping last week – changing the styles of the pictures. The internet exploded with Ghibli-style transformations of everyday photos.

Here are other styles that ChatGPT can recreate your photos:

The results are truly stunning!

Tip: Want to Mix Styles?

Yup, you can combine them! Try things like:

• “A Ghibli-style city with cyberpunk neon signs”
  
• “A Pixar-style character in a Baroque oil painting”
  
• “Editorial skincare ad in the style of an old Renaissance portrait”

Bottom line

These are the new features of OpenAI Image editing. Some new features like removing/replacing objects, changing background, and re-styling images are really impressive. 

Other functions, like creating human images or changing their feature,  leave space for improvement. Nevertheless, it is only a beginning. 

In the near future, we’ll see how AI image-generating models, like ChatGPT, will quickly improve, bring new capabilities, and broaden the horizons of visual production.