Зарегистрируйтесь сейчас для лучшей персонализированной цитаты!

I use Photoshop's AI tool every day - here are my 5 essential tips for the best results

Jun, 06, 2024 Hi-network.com
before-after

These are the before and after images of a virtual room redesign using AI. Read on to learn how it was done.

Screenshot by David Gewirtz/

It's hard to believe that Photoshop Generative Fill has only been out since January. It's become an essential part of my workflow. I use it for many projects, small and large.

I've used it to replace or create a small item in an e-commerce image for my wife's website. I've used it to slightly hack a piece of wallpaper for our family room computer to improve the foliage concentration. And I regularly use it without any parameters to clean up artifacts from images (like smudges, hair, and imaging sensor artifacts).

Also: How to use Photoshop's Generative Fill AI tool to easily transform your boring photos

When the feature first came out, I went wild with it just to see what I could get it to do. I took a simple picture of a field in front of a country highway and added a tank, a flying saucer, a bus full of onlookers, a refinery, and even an evil clown. It was cool, if silly.

But since then, I've used the feature as an adjunct to all the other Photoshop features I use daily. This week, I took a studio photo of a Raspberry Pi with an AI "hat", and added a cluster of raspberries and some mint leaves in the background.

Here's what came out of my studio. This was taken with mySony Alpha ZV-E10 mirrorless camera and my very belovedSony SEL30M35 30mm f/3.5 e-mount macro lens . Just look at that sweet, sweet depth of field!

orig-pi
David Gewirtz/

I brought that image into Photoshop and used the Healing Brush to get rid of the little bits and pieces of dust and clean up some of the various blemishes on the surface where the photo was taken.

The result was good enough. It would have made for a fine enough "hero" image for my interview with Raspberry Pi's CEO. But I thought it could benefit from a little zhush here and a little zing there. In other words, it needed raspberries.

Also: Exclusive interview with Raspberry Pi CEO: New$70 AI kit 'a watershed moment for us'

I selected the two foreground areas and asked for "scattered raspberries." I got... garbage. A random field of red droplets that only Dexter could love. It wasn't blood, not really, but it didn't look like anything that came from this Earth.

That leads me to my first tip.

1. Select only one thing to modify at a time

My next try was selecting the left front of the image and asking for "scattered raspberries." Then I selected the right front of the image, slightly letting my selection overlap the board, and again asked for "scattered raspberries." As you can see below, it came out great.

raspberries
David Gewirtz/ with Photoshop Generative Fill

But the background area was still a bit empty. I typed "scattered raspberries" into Google image search and while a lot of the images were just the fruit, and other images were of sweets, some had green leaves. I liked how that looked, so I decided to add it to my image.

My only challenge was that I didn't know what leaves to ask for. My first try asking for a bed of leaves resulted in a bad flashback to my autumn raking chores when I was a kid in New Jersey. So this time I turned to ChatGPT. I asked "what kinds of green leaves go with raspberries."

It gave me ten answers, but at the top of the list was "mint leaves." So I tried that. I selected the open area at the back of the image (overlapping a bit with the device), and asked Photoshop for a "bed of mint leaves."

As you can see above, it not only gave me a nice bed of leaves, it also put some behind the device and replicated my depth of field. And I was done.

Redesigning a room

For me, this sort of augmented studio shot is my Generative Fill bread and butter. But I want to show you a different example, so you can see its use in a far different context.

Here's my starting image. When Hurricane Matthew came blasting through my Florida town back in 2016, I was on deadline. The complete absence of local power didn't mean anything to the demands of the project I was working on. So I packed up and found the nearest hotel with power and Wi-Fi and settled in with my laptop and my VPN.

Also: The best VPN services of 2024: Expert tested and reviewed

This is the picture of the living room space in the suite I rented. I wound up having to rent a suite to get a desk, unless I wanted to drive another hour and a half farther on roads already packed with evacuees.

the-room
David Gewirtz/

It wasn't bad. It certainly served its purpose and I was able to deliver on time. But what if this wasn't a hotel room, but a room in, say, a house you want to sell? What if you wanted to show a prospective client what it might look like if it were decorated somewhat better?

Also: The best AI image generators of 2024: Tested and reviewed

Note the use of the word "somewhat". There are limitations in Adobe's Generative Fill.

The first thing I did was remove all the items in the room. The screenshot below shows that I removed everything individually. When I removed the desk and chairs (the first chair, then the second chair, then the desk), Photoshop decided to give me this bigger couch. Sometimes it's better not to ask why. It's an AI. Even so, this time, it was fine for my picture.

I also left the lamp near the drapes and the picture next to the door. The picture didn't want to remove and I had plans for that space, so I left it alone.

empty-room
Screenshot by David Gewirtz/

Next, I started adding in furniture, which brings me to my next tip. 

2. Make your selection the size and location of the object you want to fit

If you want a comfy chair, don't just draw a circle in a corner. Draw a selection around where, in your mind's eye, the object will go.

furniture-1
Screenshot by David Gewirtz/

When I got to the right side of the room, I wanted to add a TV opposite the couch. Photoshop would not let me generate only the corner of a TV. So I used Generative Expand to make the image wider, giving me a wall to the right of the door.

I asked for a TV on the wall, but it didn't like that. I got a bunch of weird images. But the weirdest image I got was when I asked for a credenza. I mean, I have no words:

credenza
Screenshot by David Gewirtz/

After many tries, it turns out "console table" worked, although it gave me a bookcase. Hey, I'll take my near wins where I can find them. "Wall TV" didn't work, but selecting the wall and asking for a "tv" did. So now I had my TV across from the couch.

I went back into Canvas Size and reduced the wall size back down to the size of the original image. I finally had the sliver of the TV that I originally wanted, showing across from the couch.

Also: I was a Copilot diehard until ChatGPT added these 5 features

The last main piece of furniture I wanted was a coffee table. So wouldn't you know it? Photoshop gave me quite a selection of credenzas, nice and tall, right in front of the couch. You gotta have a sense of humor.

I eventually tried "low center table" and got something somewhat like what I wanted. This is where that "somewhat" from earlier comes in. A big advantage of Photoshop Generative Fill is it only pulls images from training data Adobe has vetted that it has the rights to, and is allowed to share those rights with you. A big disadvantage is that it only pulls images from training data yada yada. The images are limited, which is why I got a fairly meh coffee table.

In any case, I went around the room, selecting shapes where I wanted decorative accessories to be, and asking the AI to give them to me. That brings me to...

3.  You have to play with prompts to get close to what you're looking for

Change your wording around a bunch. Try things that might not make sense. Sometimes a credenza is a coffee table to the AI. A dresser might be a bookshelf. Use a thesaurus if need be. Ask ChatGPT to give you words related to the thing you want. Mix it up and have fun. I tried different prompts if something didn't come out like I wanted, and sometimes just gave up or went in a completely different direction until I got something that looked good.

Once you feel like you're on the right track, keep that prompt for a while and try Hint#4.

4. Hit that regenerate button until you get something you like

But know that you may never get something you like from the prompt you're using.

Finally, it was done. All that was left was to add some kind of finishing touch. Sometimes an image just doesn't look complete and needs a little somethin' somethin' to make it work.

5. When in doubt, add a dog. Or a cat

I chose to add a dog (just because I liked the idea of a dog in that empty space). And here it is. You can see all the prompts on the right side of the screenshot.

final-image
Screenshot by David Gewirtz/

You can see the before and after at the very top of this article. The editing changes transformed the room from what was clearly a hotel room to more of a living room.

Have you used Adobe's AI features? What about Generative Fill? What have you done with Photoshop's AI? Let us know in the comments below.


You can follow my day-to-day project updates on social media. Be sure to subscribe to my weekly update newsletter, and follow me on Twitter/X at @DavidGewirtz, on Facebook at Facebook.com/DavidGewirtz, on Instagram at Instagram.com/DavidGewirtz, and on YouTube at YouTube.com/DavidGewirtzTV.

tag-icon Горячие метки: Искусственный интеллект 3. Инновации

Copyright © 2014-2024 Hi-Network.com | HAILIAN TECHNOLOGY CO., LIMITED | All Rights Reserved.