What is with this AI Image Look?

[amirm]

Seems like everywhere I turn, people are using AI generated images of people. How do I know? They all have this look:

Same bronzy, very warm look with that diffused background that doesn't look like a camera lens. Foreground is also aways sharp. An AI detection site says the above is from MidJourney. Maybe those always have the same look?

Regardless, it is getting way too obvious and tiring to look at. Above is from Totem website by the way. I routinely see thumbnails in youtube generated the same way.

 

[Doodski]

Seems like everywhere I turn, people are using AI generated images of people. How do I know? They all have this look:

Same bronzy, very warm look with that diffused background that doesn't look like a camera lens. Foreground is also aways sharp. An AI detection site says the above is from MidJourney. Maybe those always have the same look?

Regardless, it is getting way too obvious and tiring to look at. Above is from Totem website by the way. I routinely see thumbnails in youtube generated the same way.

It looks like a painting that is done very accurately. Around the eyes and cheeks it looks phony and the shading is off too. Like a HD image that has exaggerated sharpness and coloration.

 

[Keith_W]

Looks a bit more like Sundar Pichai to be honest.

 

[amirm]

View attachment 391718

Looks a bit more like Sundar Pichai to be honest.

It still has that "look" with the fake soft background.

 

[Mikig]

I still stand out…luckily I'm blond without a beard, without glasses and without a bronze “halo”!!!

 

[antcollinet]

View attachment 391718

Looks a bit more like Sundar Pichai to be honest.

Certainly doesn't look anything like his avatar

 

[JeffS7444]

Most recent Adobe Photoshop beta generated the attached images based on my input "female standing in forest". Kudos to Adobe for attempting to represent different ethnicities if not ages or body types, and this is the first time I've seen Adobe's results so fully clothed.

But I gotta admit, the sheer ease of generating images can be pretty entertaining.

 

[JeffS7444]

In all of these images, I see simplified textures, limited contrast, and color grading calculated to please a general audience. And bokeh, lots 'o bokeh.

 

[PierreV]

At the core, they are all trained on the same initial datasets (such as LAION 5b) and most of them rely on the same conceptual diffusion model as it as supplanted generative adversarial networks and variable auto-encoders at least for general public use.

You can tune to some extent, combine VAE and diffusion for example and, of course big players can create their own datasets and tune/censor their outputs but this won't overcome the AI look without fundamental algorithmic changes.

And, for economic reasons, what we get is mostly the "fast-food" version of image generation. Relatively easy to train and to serve for the general public's entertainment.

 

[mhardy6647]

conceptual diffusion...

you know, I am not sure that I am glad (for him) -- or deeply sorry that Frank Zappa isn't with us on this plane anymore.
He'd have had pithy observations, I am sure, vis-a-vis AI.

Oh.
Speaking of which...
Has any of all y'all asked your favorite AI... umm... thing to whip up an affable, gauzily plausible image of Mr. Zappa?

(not AI -- source: https://wiki.killuglyradio.com/wiki/Marshmoff)

 

[DLS79]

Same bronzy, very warm look with that diffused background that doesn't look like a camera lens. Foreground is also aways sharp. An AI detection site says the above is from MidJourney. Maybe those always have the same look?

With everything Ai imagery, text, etc output is always dependent on the training set.

Thus I'd be willing bet the training set had a lot of studio head shots in it. Then since the general public drools all over the hdr look, the algorithm probably tries to push dynamic range as far as it can.


One of the reason they are becoming so common is because people don't have to buy stock photo libraries, or pay royalties for ai generated images.

 

[Brian Hall]

AI is getting very good at making videos from pictures:

 

[Brian Hall]

 

[Kremmen]

Thank you for linking to these videos Brian. AI may be a concern for many, but when used in this fashion it can create some truly beautiful and poignant moments. I like the first one the most.

 

[Brian Hall]

Thank you for linking to these videos Brian. AI may be a concern for many, but when used in this fashion it can create some truly beautiful and poignant moments. I like the first one the most.

I agree. The segment with Marilyn Monroe (not the one with the blowing dress) came out very well.

 

[Titurel]

It still has that "look" with the fake soft background.

Most portrait photography (these days) goes for exactly that look - focus on the eyelashes and adjust the aperture (typically wide towards wide open) for the desired depth of field; adjust ASA and shutter speed to affect proper exposure; if necessary, sharpen the face in post, leaving the background alone. There you have it, sharp eyes and soft background. In the image you posted, focus is already falling off on the couch back behind the subject's shoulders. The eyes may have been "dicked" with using an "A.I." type sharpener (e.g., Topaz), but I've seen shots I know were shot on film, that look more phony. If this photo's really a fake, it's a pretty damn good one.

 

[Brian Hall]

4K | BEAUTIES IN SPACE | AI Generated - AI Film 4K​

 

[Brian Hall]

Near Space | Sci-Fi AI generated future video​

 

[Brian Hall]

They are getting very close to better than reality.

 

[Doodski]

They are getting very close to better than reality.

Did you request those videos from the ChatGTP thingy?