What AI can actually do well - a brief Photoshop tutorial for using generation to make good mockups

PenguinPower
Valued Contributor III

AI may have a long way to go, but it can do somethings well... If you've got access to Photoshop and have generation credits, you can make high quality and unique mockup images quickly and efficiently. The images have plenty of resolution for this use. 
My basic process. 
1) Download your mockup image of choice
2) Remove the background (I have saved selection files to make this process a single step)
3) Move your item to a new position in the frame, unless you'd like it to be dead center.. 
4) Select the inverse 
5) Type in your request for the background you wish to generate. 
6) Bring the mockup image layer to the top
7) Adjust your mockup image - the generated background will include a bit of your mockup image duplicated around its edge. With mugs I found I this would often look odd.. either make the mug image larger to cover or use the rubber stamp to fix any strange edges. With the beer glasses, for the most part they looked fine and could usually be left alone. If there were oddities on the edges, adjusting the size of the mockup layer to cover was sufficient. 
Add any extras - for the most part lighting, shadows and reflections will be taken care of by the generation. For a little added realism I chose to add "steam" to many of the mugs. and shading behind beer glasses. 
9) flatten your image, resize if desired. 
10) save as a jpg (or png.. but jpgs will upload a lot faster). 
11) upload your new mockup.
As long a you can keep thinking up things to type into the generation window, you can keep getting new mockups.. I managed to do around 200 products in just a few days. 
I'm happy to answer questions about specifics... if I've left anything out or anything seems unclear.
here's a sample of what I was able to plow through.
Screenshot 2024-01-18 at 6.16.25 PM copy.jpgScreenshot 2024-01-18 at 6.17.41 PM copy.jpg

 

22 REPLIES 22

Windy
Honored Contributor II

Good info!

I also do Postcrossing!


PenguinPower
Valued Contributor III

I've also done a screen recording of the process -
https://drive.google.com/file/d/1WnAM8dj0S1z2IrS4mgl2B6rFAFU4M4ei/view?usp=sharing&t=10


(there isn't any sound!)

Windy
Honored Contributor II

Brilliant! I had tried the photoshop generative fill a few times with lackluster results--that's been my only foray into AI. After watching your video, now I see how prompts can be used to good advantage. I had no idea you could include all that, even up to the out of focus background. Thanks for this.

I also do Postcrossing!


PenguinPower
Valued Contributor III

It helps to remember that Adobe has trained their AI with their own stock library - phrases that will result in good results in the search engine for the stock library will usually result in better images generated by the AI. Unlike searching for stock though, you can string terms together to get more specific results. Searching that way can cut down on your results too much... but it works with the AI, to a degree.. I've found it gets confused when you ask for multiple subjects -so asking it for a burger and fries tends to confuse it. It will give me a burger or fries, but not both together, unless I ask it for a "hamburger meal" - and how it interprets that can be... interesting... but multiple descriptors work fine. If I ask it to put that burger on  "a  table in a garden in the springtime near sunrise" (it doesn't care about punctuation), it will do pretty well at creating all of that, and give me a tabletop in the foreground and some flowers with muted light in the background. I ask for a lot of blurred backgrounds... leaving the detail up to the imagination when possible is usually better. I means that your brain fills it in rather than there being obvious AI wonkiness. 

Thanks for sharing that! I've had middling luck with AI in PS so far. Definitely seems like more description really helps.

__________________
Claire of Rocklawn Arts

Connie
Honored Contributor

Wow, that is really good! Thank you for sharing your workflow.

KeegansCreation
Honored Contributor

Can confirm. I have done this many times since Photoshop AI came out of beta.

KeeganCreations

Windy
Honored Contributor II

I am totally going to have WAY too much fun now that I know how to do this. Thanks @PenguinPower 

I also do Postcrossing!


PenguinPower
Valued Contributor III

Awesome 😀
I gave it more of a challenge and tried back packs. I was pleasantly surprised that it somehow can figure out that it's supposed to put a person into it.. They took a little handwork to fix some weirdness/edges, but they came out quite well, I think.

I did however have to go back to stock for this one... the AI just cannot handle the detail of a wheelchair without going all goonie and wobbly.. 

excellent results

 

KeeganCreations

Windy
Honored Contributor II

Here's a bearded fisherman on a boat. I guess the blue cooler tells us he's on a boat, though I had actually been hoping for a longer view--with some of the boat in view. Originally I asked simply for a fisherman, but one of the AI options came up with a beard, so then I asked for a bearded fisherman, and I liked those results better. I made a separate selection above the mug and asked for steam, but I really need to learn how to just make some steam without the AI assistance.  

Probably, so that I do not waste credits, I should do my first search as a simple google search, so that images google returns, will give me an idea of what kinds of elements I may wish to include in my AI prompt.

I also do Postcrossing!


PenguinPower
Valued Contributor III

Making steam is easy.. make a new layer above your mug. Take a brush and draw a few white squiggles - it’s ok that they are silly and cartoonish. Run a Gaussian blur on the layer. Use the smudge tool to swirl and shape to your liking. Run a second blur if necessary. Adjust the layer opacity if desired. Use the eraser to clean up anything that has gone below the mug rim. Voila steam. 

Windy
Honored Contributor II

Thanks for taking the time to tell me how to make steam. I finally had a chance to give it a try. And I'll keep working on this technique......and also the AI prompts, too. 

I also do Postcrossing!


Connie
Honored Contributor

Wow, you've been busy! How do you know what to type for the prompts? can you give a few examples, please?

Do you still have unlimited generation credits with Photoshop? Now they put a limit on it, so it would take several months to get all these done, unless EVERY image is usable.

PenguinPower
Valued Contributor III

I don't have unlimited generation credits, but I have a subscription to the full creative cloud package, though not the stock library - I only use the free images, but anyway, I have a lot... It looks like I get 1,000 credits a month?, even working flat out like that I only used about 600 of them... and they roll over, so I don't think I'm going to run short. My "hit" rate was pretty good. I wasn't generating multiple times without getting something useable that often. 

I'm finding that being "medium" descriptive is usually best.. too little and it doesn't really know what to do.. too much and it gets confused or It will just ignore things, but it's still usually better to give it too much than not enough. I also don't try for items that are overly detailed... asking for a blurred background has been working well for me. Most of what I've worked on recently has been glassware. AI still only has a vague idea what a kitchen should really look like.. but blurry it's all good. 
Here's a few examples.  
The prompt aren't saved when you do them directly in Photoshop, but it would have been something like "a plate of dumplings on a countertop in a rustic kitchen. Background out of focus." It might have just been dumplings though.. often asking for a plate would result in the plate being stuck under the mug/glass.. 
botanicQbeercover.jpgbotanicQbeercover.jpg

This one would have been "a stack of oatmeal raisin cookies on an artistic background. Background out of focus" medfigFmugcover.jpg

This was " a poker game on a coffee table in a modern living room. Background out of focus"medfloralObeercover.jpg

Connie
Honored Contributor

Oh, I se. The full plan gets 1000 credits a month. The Photography only plan ($10 a month) only gets 250 credits. Generating 3 versions at a time, means I would only be able to experiment 83 times. and it usually takes me several tries to get a usable image.

PenguinPower
Valued Contributor III

Looks like our corporate plan doesn't get any more than that - I looked today and it was also 1,000. 

PenguinPower
Valued Contributor III

Now it is getting better at more complex things... These took some creativity and some work... So a little background here. I live in a large city, a large city with a lot of dog owners.. a lot of dog owners that don't feel the need to scoop.. It's not all that unusual a topic of local conversation 😪.. I already do a decent business in stickers that either ask folks to not use your garbage can for poop bags, tell people you don't mind if they use your garbage can for poop bags, tell people you don't mind, but for pity's sake.. make sure it's the garbage can and not the recycling bin... So decided to give a go at making some please scoop signs too. I'm good at some kinds of graphics but I'm not a painter and I wanted a "Poop Fairy".. I was able with a bit of work to get two very funny images that are plenty large enough for 12x12 yard signs... I couldn't not however mention poop in the prompts.. it would run afoul of the censors. I made these in the Firefly app. Generation inside photoshop wasn't cutting it when I had nothing to start with. 

Here's no 1 - I had to add the poops in by hand. The prompt for this one was "a dog with wings wearing a pink tutu and waving a fairy wand" 

and no 2 - this took a good number of tries to get a person that did not have a messed up face or messed up hands.. or a chair that wasn't all wonky - or was anything like I was actually asking for, but in the end I got something I was satisifed with. (the bug seems to have disappeared the yard sign.. but I made a plastic door sign too) The prompt for this one was "a sloppy looking overweight old man with tattoos, wings and wild gray hair, wearing a dirty sleeveless undershirt and a pink tutu, sitting in a lounge chair by a pool in a rundown motel" - I never did get it to make exactly what I wanted - which was a pool and surroundings like you'd find in an old 50's motor court that had seen better days... I tried a lot of descriptors, rundown, shabby, seamy (the censor nixed this one), 'seen better days/times"..  but I think that was just a bit too far a reach for it.. so the poop fairy got a little better retirement than I'd imagined for him 😆

Those are hilarious signs.

KeeganCreations

Windy
Honored Contributor II

Oh, how I love these Poop Fairy signs!!! LOLOLOLOL! 

I also do Postcrossing!


PenguinPower
Valued Contributor III

PS - at my actual work we have a corporate subscription... it says it includes unlimited stock images... I don't actually know how many generation credits that kind of account has - I work in healthcare/sciences so using generation would be an immense no-no... so I've never had a reason to need to know, but I can take a look and see. 

Windy
Honored Contributor II

Today I asked the AI for "with cookies in a treehouse". So it put my mug in front of a tree with two chocolate chip cookies stuck to the bark above the mug.  But many of the results I am getting are pretty darn good!

I also do Postcrossing!