I'm currently only looking at it at my phone but... yeah, I think you're actually right. In which case you're most likely right about the catdog too. That is actually impressive. And a little scary.

Ok, at first glance I didn't see anything suspicious but looking again, that one does look weird. The hands... And more importantly, the entire cat is slightly different. So that looks like the case of what I've been talking about - the entire image was re-generated, not just the erased parts. Which is what I'm aware the current tools work like. (But again, feel free to prove me wrong). In this catdog picture, the dog head and the entire background seems completely unchanged though. So either it was not AI, or it was some combination of AI and regular photoshopping, or there's a tool out there, that can do this but the question then is why it wasn't used with the paparazzi cat picture.

>You really think it's easier to find a weird blanket from another picture and match the lighting and everything?

Absolutely. Look at the area you'd have to erase and perfectly match. Three different surfaces. Clone stamping would be a pain and content-aware "AI" tools are a hit or miss and almost surely would fail here. The lighting is neutral so if the other image was similar, no matching had to be done, just some drop shadow. The only odd thing may be that they didn't reuse the blanket from the original image. But maybe the pillow was part of the cat picture.

>AI tools do have inpainting functionality where you erase a part of a picture and then describe what you want the AI to generate in the empty spot.

I am aware, but from what I've seen, even such tools re-generate the entire image. But the tech is evolving so fast that I may be wrong. So if you can show me a particular tool that's capable of this, I'll give it to you that AI may have been involved. But I still don't see why it would have to be AI, I think you're reaching.

The odd piece of blanket serves the purpose of hiding the dog's hind legs. Much easier than trying to match the surrounding background.

Not sure about the toes, could be a clone-stamping accident in an attempt to stretch the cat's leg to, again, cover the dog's front legs. Or the cat just has a strange looking paw with extra "fold".

I don't think there's currently an AI tool that could edit a photo in this way with such a level of precision and specificity and while leaving the rest of the image intact. I mean Photoshop has content-aware fill which does use some level of machine-learning and that can be used to erase an object from a scene, but it can't replace half of a dog with half of a cat.

All you need is an ND filter (which is basically photographic equivalent of non-polarized sunglasses) that allows you to shoot long-exposures during the day without overexposing the sensor.

It looks unreal because just like any other long-exposure photo, you can't see such effect with the naked eye, and additionally, people are mostly only used to long-exposure photos taken after dark (where you don't need a darkening filter because there's not enough light already).

So while you can sort of argue that the photo is fake (but that's a very slippery slope because no photo perfectly captures reality, not to mention neither do our eyes and brains), it is absolutely possible to shoot this (both on film and digital cameras) without any post-processing.

I don't know where you live but up until a few years ago, every single lamp in my city had exactly this color, because they were sodium-vapor lamps. They've been replacing them with LED lights but still plenty of sodium lamps remain. If you do an image search, you'll find plenty of photos that look almost identical to the pixelart. One example: https://commons.wikimedia.org/wiki/File:High_Pressure_Sodium_Lamps.JPG

Star Child