AI-generated images: friend or foe?

[libordeath.com]

I'd like to learn more about the impact that AI-generated images are having on prepress workflows: how frequently customers are submitting them; what kinds of production problems they can cause; whether they create legal issues around privacy, ownership, and copyright. How are shops dealing with the AI-generated graphic imagery that they are being asked to reproduce?

 

[tngcas]

The biggest thing we are seeing is that people are submitting chatGPT generated images that they spend a lot of time working with chatGPT to create. IE. I got one the other day from a salon where they used chatGPT to create a flyer with pricing on it for all their services.

ChatGPT refuses to give you high resolution files though even when specifically asked for it (I've tested this). If you tell chatGPT I need this file to have 3300 pixels with 300ppi it will still only give you at most an image with 96ppi (and nowhere near 3300 pixels).

This causes issues because customers spent a lot of time on getting the wording right and it's now their "file" is in a completely un-editable format (png) and low resolution with no easy way for the customer to fix and/or recreate the document. The image looks good to them on their screen but isn't really printable at a decent quality.

Having said that, I have found a few ai's (like envato elements) that seem to be able to upscale (within reason) chatGPTs low resolution files into better quality prints but that's not really publicly accessible for customers yet. I'd love a way to be able to send our customers to an AI that they can submit their chatGPT images to and have them upscaled to a print quality version. But alas that day is not here yet.

 

[smalloffsetpressexperts]

GROK works decent and is a great source of idea's and how to's. I made this for a project

 

[JeffJ]

I have created logos for some of my customers. The AI generated a png. file. Which is as you know is not a printable file.
I searched and found an app that converted it into a higher resolution file that worked pretty good.
Bubbi.app

 

[kslight]

I have a couple customers that employ AI at the design stage. They tend to look nightmarish, in all the cliche generative AI ways (extra digits, melting faces, impossible structures, etc). But if they sign off on it that’s their problem. When I have customers requesting our design services I inform them of stock image licensing fees if required…it is much less time consuming to find print quality images that get in the ballpark of what we need than to fool around in the slop.

 

[gordo]

Printers should always use A.I. - i.e. Actual Intelligence

 

[namelessentity]

 

[smalloffsetpressexperts]

Guys embrace technowledgy like you embraced copiers and called them a press and poo poo'd offset, which explains my avatar Offset has been mostly ghosted. A.I. is the future

 

[wonderings]

We don't use it for creating images in our design work, we do use it for adding bleed, removing pieces and other little things like that. Saves a ton of time and very versatile. We did have a design house submit a file for a banner for a credit union and they had taken an ai image from adobe stock. Client did not notice this not their digital proofs and refused the product. It did look very weird in the face when you stopped and actually looked at it. So I tell my staff to use it for fix up stuff, but to avoid it for creating anything from scratch.

 

[TJPrinter]

We don't use it for creating images in our design work, we do use it for adding bleed,

I'm curious about what AI you're using to create the missing bleed. Does it do a better job than Pitstop?

 

[wonderings]

I'm curious about what AI you're using to create the missing bleed. Does it do a better job than Pitstop?

Indesign and Photoshop both have AI in them now and you simply use Generative Expand. I don't put any prompts in, just let it continue on with the image.

Below is a quick example in where you will see just a random picture placed in an Indesign document. Now if I drag out the box that holds the image to the bleed edges you will see nothing as there is no more image to show. There is a small tool bar floating above the image, when you have an image selected it will change the option to "Generative Expand". You select that and are moved to a new small window where you can type in a prompt of what you want or just leave it black and have it generate.

What the generative expand then does is extend the image while making it looks seamless and even gives you 3 options to choose from. We don't use pitstop and this is just a simple time saver. I am assuming if Pitstop is not using AI that it is just automating the manual way of adding bleed by flipping an image and butting it up against the main image. This has limitations, AI is generating new artwork so if you have text or anything else close to the edge you are not going to see that in the bleed.

 

[narseman]

Having said that, I have found a few ai's (like envato elements) that seem to be able to upscale (within reason) chatGPTs low resolution

Explain the problem, offer to take care of it, for a small upcharge, of course.

 

[tngcas]

Indesign and Photoshop both have AI in them now and you simply use Generative Expand. I don't put any prompts in, just let it continue on with the image.

I'm gonna have to play around with that.

 

[smalloffsetpressexperts]

This one IMHO is 4.5 stars. I have CCloud also and now pay for credits for the generators I pick

 

[wonderings]

This one IMHO is 4.5 stars. I have CCloud also and now pay for credits for the generators I pick


View attachment 294645

I use ChatGPT Plus myself, have a subscription to that service and find it helpful for so many tasks. I took a prompt and tried it with both Grok and ChatGPT. I thought Grok would be better in this case but ChatGPT was actually closer to my prompt and the face looks more accurate.

Prompt: Tom Cruise riding a tricycle on top of a hot air balloon that is floating over snow capped mountains. The image is a close up of his face with the top of the balloon behind him and them mountains further back.


Grok image:

ChatGPT:

 

[ReproElectroProspero]

Here's Gemini Nano Banana's output from your prompt:

 

[pippip]

Mad the difference, I wonder how much the users past interactions with each ai is influencing the outcomes.

Scary really

 

[ReproElectroProspero]

Mad the difference, I wonder how much the users past interactions with each ai is influencing the outcomes.

Scary really

It truly is. As someone that's made a living over the last 20 years as a graphic designer / layout artist - I'm actively searching for alternative skills to learn. The next few generations of these things can and will make a lot of what we do obsolete. Imagine being able to prompt "a blue/green color letter size flyer advertising my business and our year-end 20% off sale.", with the generator being smart enough to use info from the client website, find and use vector logos/brand assets, and add/edit text with zero "ai slop" being apparent.

We're like a year away from this imho. Think back to that first AI generated video of will smith eating spaghetti. In two/three years we went from "obviously fake" to "almost perfectly realistic".

 

[tngcas]

We're like a year away from this imho.

Agreed. The winners are going to be the people who learn how to write effective prompts.
I also found chatGPT to be better at this than grok (so far) but the envato elements ai is way better than both of them.

 

[pippip]

It truly is. As someone that's made a living over the last 20 years as a graphic designer / layout artist - I'm actively searching for alternative skills to learn. The next few generations of these things can and will make a lot of what we do obsolete. Imagine being able to prompt "a blue/green color letter size flyer advertising my business and our year-end 20% off sale.", with the generator being smart enough to use info from the client website, find and use vector logos/brand assets, and add/edit text with zero "ai slop" being apparent.

We're like a year away from this imho. Think back to that first AI generated video of will smith eating spaghetti. In two/three years we went from "obviously fake" to "almost perfectly realistic".

View attachment 294655

The Will Smith eating spaghetti is actually the unofficial benchmark for ai, used by the industry to compare new ai engines and models.