Jo Ann,
even though I don't post much on the subject myself, I'd like to thank you once for your "AI obsession".
You always dig out here highly exciting and interesting aspects on the subject. It's fun to read. 

I can confirm this. I have started to prompt people with stable diffusion and it feels like there are maybe 10 -20 faces for every ethnicity and then the ai �decorates� them with different hair styles and clothes.

In ai groups I read they get around that problem by creating people through mixing famous actors or personalities. This is apparently also a way to create a consistent  �character�, that they then can build a world or stories around.

I am sure in the next generation of ai, there will be more people options and also the options to create real characters or avatars that you can use for your work.

eta

One thing I am experimenting with is actually giving the model a name and a little story - a picture of anna the nurse who likes to write murder mystery novels...paul, the teacher who plays in a rock band...etc...

Most of the time it does not work but sometimes I get some very interesting results with completely new faces and the consistency is increasing.

So you don't need famous people. Instead, ask yourself what are the visual character attributes of these famous people? How would you describe the vibe they give you?

The ais can do a lot more than we see, the extreme similarities come from everyone copying the same prompts.

Just like there is more than one tropical beach, but hardly anyone makes the effort to prompt something really new.

eta2

If I get a series that I like I first check how many similar ai images already exist on Adobe. If many ports have basically the same image, I might discard the series or just upload one metoo test file.

Then I will revise the prompt to get a different look. Even if it is just a subtle change, I think buyers will appreciate if you make an effort to try new ideas.

I think agencies will also start selling avatar packages �set of office people�, �set of happy children�, that will keep their visuals while prompting different scenes, ready to use for projects.

Might become like in computer games.

Quote from: Jo Ann Snover on August 08, 2023, 15:38

Quote from: cascoly on August 08, 2023, 14:01

while looking at SS ai gen'd images (which they earlier said couldn't be submitted) i found this one with terribly rendered faces

https://www.shutterstock.com/de/image-generated/wet-plate-photo-indian-village-where-2329686851

No wonder in the earnings call that SS execs mentioned that the quality of photo-reallistic AI needed to improve before customers would download!! That's not even usable as a flood-damaged or fire-damaged photo. Embarrassingly bad.

I don't know what you mean. it's all good and fine

https://www.shutterstock.com/image-generated/bokeh-photo-red-yoda-2310850257

Oh . my . deity-of-choice!!!

I see your Yoda and raise you a young man writing a speech

https://www.shutterstock.com/image-generated/advertising-product-photo-young-male-writing-2344753047

I looked at the first few pages (sorted by most recent) of the AI generated stuff and the comment from last week's SS earnings call now makes perfect sense. The gist was that lots of customers were trying the AI tools but not many were being downloaded because of quality issues.

Even the illustrations are a disaster

https://www.shutterstock.com/image-generated/pupils-actively-engaged-health-activities-vibrant-2344797389

I didn't see anything even close to usable. Nothing.

https://www.shutterstock.com/ai-image-generator

Shutterstock customers generate the images that show up in their acknowledged AI generated collection

https://www.shutterstock.com/search?image_type=generated&sort=newest

From the top link:

"Shutterstock users looking for AI-generated images in our core library can use the AI-generated image filter on the Image search results page. All AI-generated images displayed has been generated using our Shutterstock AI image generator. We are still unable to accept AI-generated image submissions from contributors due to concerns regarding the safety and licensing of content created using other generative tools and models."

They have bucket loads of AI generated images that have been uploaded by contributors that aren't marked as such and that their inspection process didn't catch

They say that millions of images were prompted...where are they? I can always just find a few thousand if I do a search and say i want ai only.

Two different images, two different contributors - one premium, one standard collection. Both the same size. Both are AI, but they are uncannily similar

.

In addition to the issues of genAI repeating itself, it's hard to see why one of them is $249 and the other regular subscription.

Two different images, two different contributors - one premium, one standard collection. Both the same size. Both are AI, but they are uncannily similar

I checked these out and they are both by the same contributor, Rafa Fernandez, but the Premium one is submitted via Addictive Stock, an agency of Spanish "photographers".

https://addictivestock.com/item/460136

Two different images, two different contributors - one premium, one standard collection. Both the same size. Both are AI, but they are uncannily similar

.

In addition to the issues of genAI repeating itself, it's hard to see why one of them is $249 and the other regular subscription.

When you let AI create images with Midjourney you get 4 images and you have the option to have the AI recreate a single image as "subtile" variation or a  "strong" variation" . The subtile variation will give you images like this. This looks very much like a subtile variation of the same image.
Unless you have a pro or mega plan all images are visible to everyone on the discord server and everyone can make variations of your images. So this is most likely either the same "contributor" using two different accounts or someoen who is so lazy that he cannot even be bothered to write a simple promt himself and just makes variations of someone else's images on Midjourney. (Edit: Ah, I see someone aready wrote that this is from the same contributor)
Either way I do not think this is the AI just creating two extremely similar images by chance, but a deliberate variation of the same image.

Oh . my . deity-of-choice!!!
...

i  read a post from a woman who said she didnt want her kids to blaspheme xtianity, so she chose to teach them to say  "Buddha" when the hit their thumb w a hammer. {much better to curse someone else's god}

i learned recently that our responses like that get hard wired & not processed by the normal brain language areas, so it's a difficult habit to break.

I asked SS support (why did i bother?) if i could submit images made w their generator and they said definitely no

So, what is going on, i still am not sure i understand! They are building a crowd-sourced AI image collection, but made by customers testing their generator and not by contributors like you that know what they are doing with AI. Sounds weird as most people will just prompt a random thing just to test the generator and will probably never create something useful.

On the other hand they save a lot of money building the collection this way, because if they accepted your generations then they would have to pay you!