Background information

Photoshop: Adobe’s new AI image generator can do a lot – but it can’t do everything well

Michelle Brändle
25-5-2023
Translation: Megan Cornish

You can now use text instructions to remove or add objects to your images. Adobe has built an AI tool within the beta version of Photoshop to do just that.

Adobe has built an image generator into Photoshop. With text instructions – it calls them prompts – you can give your pictures a different background, remove annoying tourists or add objects. This is all possible with the new beta version of Photoshop. It’s available for free if you already have an Adobe subscription. Read on to find out what you can do with it.

Generative Fill and Remove Tool

You can use the Remove Tool – which you can now find alongside all the other tools in the toolbar with a plaster icon – to roughly highlight the unwanted object. Photoshop will then automatically remove it with a single click.

I roughly highlight two people in the background…
I roughly highlight two people in the background…
Source: Michelle Brändle
…and, with just one click on «Remove Tool», they disappear from the photo.
…and, with just one click on «Remove Tool», they disappear from the photo.
Source: Michelle Brändle

Adobe made the AI image generator available for testing in March. It’s in the so-called Firefly family, which I reported on for its launch. Now the image generator’s «Generative Fill» feature is integrated directly in Photoshop.

To insert objects and backgrounds via the AI image generator, roughly highlight the area you want to change with the selection brush. The «Generative Fill» tool will appear automatically. Then you use a prompt to select what the software should add. This currently only works in English.

I added an erupting volcano to my holiday picture by selecting «a volcano erupting». The AI then gives me three suggestions. You can see the coolest option here.

The volcano looks really cool.
The volcano looks really cool.
Source: Michelle Brändle

A beach in Mexico was virtually deserted. I wanted to add an ice cream cart. The image generator produces a strange result from my prompt.

There’s now an ice cream truck on the previously empty beach. Or something like that.
There’s now an ice cream truck on the previously empty beach. Or something like that.
Source: Michelle Brändle

I tried other prompts, but I wasn’t happy with the results. Neither «an ice cream selling car» nor «a car with ice cream in it» nor «a bus with a person who sells ice cream» works. The AI probably doesn’t know what good ice cream looks like. Or a suitable car.

With the added human, the result looks even more surreal.
With the added human, the result looks even more surreal.
Source: Michelle Brändle

The program uses its own database to create the objects and backgrounds. This avoids any copyright infringements.

I can also do useful things with the AI function, such as removing the background or converting an image from landscape to portrait. Both work quickly and easily. The fact that the AI cuts out my hair so accurately is impressive. And I can touch up anything I don’t like later. The tool works with layer masks.

My hair is cut out to the finest detail.
My hair is cut out to the finest detail.
Source: Michelle Brändle
The previously landscape format image is now in portrait format with more sky.
The previously landscape format image is now in portrait format with more sky.
Source: Michelle Brändle

Image corrections: preset adjustments

Normally, you have to manually adjust an image’s brightness, hue, tone curves and so on if you want to achieve special effects. There are new so-called preset adjustments. They work like filters that are placed over the image as layer masks. For example, if you want a sepia effect on your image, the program uses your selection to create two layers in a «Sepia Colour Repair» layer folder: one for the colour balance towards reddish-brown and one for black and white. This has the advantage that you can adjust the individual levels individually.

My harbour photo gains a sepia effect at the touch of a button.
My harbour photo gains a sepia effect at the touch of a button.
Source: Michelle Brändle

Quick selection when editing images: contextual taskbar

This small menu bar is displayed on the desktop and can be moved around. It shows the most common next actions when editing images, so you always have the most logical steps at your fingertips.

The small menu bar adapts automatically, depending on the stage.
The small menu bar adapts automatically, depending on the stage.
Source: Michelle Brändle

Gradients with a preview function

You’ll find a small but very useful feature when creating a gradient. Previously, you had to laboriously try out how the set gradient would look on your image. Now you can see a preview while you’re creating, so you no longer have to undo elements that have been set.

The gradient is displayed as a preview.
The gradient is displayed as a preview.
Source: Michelle Brändle

Quick feedback for projects: creating links

In my opinion, one of the best new features. When I work on a job, I often have dozens of screenshots or preliminary versions that I have to go back and change after consultation. Well there’s a simple solution. I create a link for my file and use it to comment on my projects. With the quick feedback, I can adapt my work quickly and easily.

This works by first saving the project, then clicking «File > Approve for review > Give project a name». A link will then be created. This will take a moment, but in the meantime you can get back to your project. You can share the link or only release it to certain people if you wish.

 Creating a link from the project will help you get feedback faster in the future.
Creating a link from the project will help you get feedback faster in the future.
Source: Michelle Brändle

20 people like this article


These articles might also interest you

  • Background information

    From horsies to horror – creating fantasy landscapes with AI system DALL-E

    by Michelle Brändle

  • Background information

    Why automated AI detectors don’t work

    by David Lee

  • Background information

    "Adobe Firefly": The web app now generates AI videos

    by Michelle Brändle

Comments

Avatar