ChatGPT’s image generator is similar to Photoshop

DALL-E, OpenAI’s image generator, now allows you to partially modify an image, to refine a creation assisted by artificial intelligence. A major step forward for artistic modifications.

In the relatively near future, will a complex tool like Photoshop be obsolete? The rise of generative artificial intelligence makes this avenue credible, while creating and modifying an image has never seemed so accessible. Tools like Photoshop retain creative finesse, since they make it easier to impose a vision, but generative AI produces content that is often more than satisfactory, in just a few seconds.

On April 3, 2024, OpenAI announced the launch of an image editing tool, which it integrates directly with ChatGPT and Dall-E. How it works ?

A brush to select a place to modify

Let’s be clear, OpenAI’s new image editing tool is infinitely less comprehensive than Photoshop. Where Adobe offers a hundred tools, ChatGPT offers… just one.

We generate an image with a sentence (“Generates an image of a person on the street in New York, walking in Time Square. He has a blue t-shirt and a cap” for example), then open it in full screen. A brush allows you to access the editor, then surround an area to modify.

This small logo allows you to modify an image.
This small logo allows you to modify an image. // Source: Numerama

Unlike Photoshop, Pixelmator or Affinity, ChatGPT does not allow the user to edit the image themselves.

When you surround an area, you must write a new prompt. In particular, you can circle the t-shirt and ask to replace it with a suit, the AI ​​then keeps the original image, but changes the surrounded part.

The different versions appear on the right.The different versions appear on the right.
The different versions appear on the right. // Source: Numerama

Before OpenAI rolled out this change, each new sentence required Dall-E to regenerate an image from scratch, preventing further development of a project (take it or leave it). The new tool makes creating a tailor-made idea possible, although it still requires trusting the machine.

Unable to import an image from your photo library

Screenshot 2024-04-04 at 11.53.10Screenshot 2024-04-04 at 11.53.10

Another limitation: the OpenAI generator can only modify images generated by Dall-E. Importing a photo leads to nothing, it is impossible to circle the sky and request a change. It’s a shame, since it could give new impetus to ChatGPT, transforming it into a real editing tool.

How to explain this limitation? OpenAI is undoubtedly cautious, but could unlock this possibility in the future. Nothing technically prevents it, especially since ChatGPT claims to be capable of it. But when we ask it to make a modification, it generates an image that looks like the one we imported, but which is completely artificial.

On the Adobe side, the creator of Photoshop integrates its own generative AI into its Firefly software. Firefly can generate content or modify an element of the image from a sentence, like Dall-E. But Firefly can also retouch real photos, which leaves Adobe with a real lead for the moment. For how long ?

Do you want to know everything about the mobility of tomorrow, from electric cars to e-bikes? Subscribe now to our Watt Else newsletter!

Source link -100