An AI expert tested Google DeepMind's free text-based image editing feature. Here are his findings

«'Conversational editing' will gradually become mainstream,» noted Oleksiy Minkov, adding that most likely it will become the new norm in the near future. «Craft editing with pens in Photoshop will be a thing of the past,» he added and shared his experiences.

How to use the feature

You can edit images for free using Google DeepMind’s text prompts in Google AI Studio. To do this:

• choose the Gemini 2.0 Flash Experimental model,
• upload an image,
• Write a prompt on what to change on it.

The expert noted that, unlike other services, the peculiarity of the Google DeepMind function is that in one window there is the possibility of interaction through a common prompt and for various tasks, from the background to changing or adding objects, without even having to select areas in the image.

Colorization of a black and white photo

Oleksiy uploaded a black-and-white photo of Lviv from 1964 and simply asked to make it color with text. After 7 seconds, he received an edited color photo.

Head of Product Roman Astafyev added in the comments to Alexey’s post that he also tried to use this function, but his results were not satisfactory. «It doesn’t do something, it gives an error after processing the photo,» Roman said. According to him, the product downloaded a random black-and-white photo from Google, which depicts people or architecture. In addition, he noticed that if you look at the sample on the screen, the photo became worse. «The face and body are blurred. It looks like it wasn’t colorized, but the generative engine painted the photo from scratch,» Roman shared his impressions.

And Maria Khomenko couldn’t colorize the photo at all. She shared a screenshot of the disappointing results of communicating with the AI ​​model.

Add an object

Then Alexey Minakov uploaded an image of the table and simply asked for flowers to be added to it in text. Here’s what happened.

Change background

Next, Oleksiy uploaded a photo of himself and simply asked with text to change the background to space.

Instead of a conclusion

Oleksiy Minakov emphasized that this is an experimental launch of the feature for testing. «Therefore, you will receive the edited image in a non-high resolution extension. We are waiting for the full release,» he added.

According to him, it is interesting that the image editing is not done by a separate model for generating images Imagine 3, but by the multimodal model Gemini. That is why he advises not to limit yourself to using only one AI tool like ChatGPT, but to be open to other alternative services. According to Oleksiy, such an approach not only diversifies risks, but also provides access to special opportunities.

«OpenAI is just planning to launch something similar in ChatGPT in the near future,» the expert recalls.

Agreeing that in AI racing, most of the functions are similar for all manufacturers, Oleksiy still noted that there are certain differences that are worth paying attention to. «Gemini, for example, has some capabilities that ChatGPT does not have, in particular, extracting abstracts from YouTube videos,» he added, explaining that diversification in this case is no longer about buying paid packages, but rather about having several AI services with their free capabilities in the arsenal.

You can use the service at the link .

Read the country’s main IT news in our Telegram

On the topic

Read the country’s main IT news in our Telegram

An AI expert tested Grok. What are its features and how does the chatbot differ from ChatGPT, Gemini, and Claude?

On the topic

An AI expert tested Grok. What are its features and how does the chatbot differ from ChatGPT, Gemini, and Claude?

12 powerful free AI tools tested by an experienced Product Owner

On the topic

12 powerful free AI tools tested by an experienced Product Owner

AI researcher and co-founder of OpenAI Andrey Karpaty tested the Grok 3 Mask: here are his conclusions

On the topic

AI researcher and OpenAI co-founder Andrey Karpaty tested the Grok 3 Mask: here are his conclusions