Finally AI upscalers no longer suck now...
It's amazing what they're capable of when they train on the exact same fucking source image in high res
There is literally no way to get around the fact that any upscaler will have to invent a new image.
>>106858189>plastic skinit's shit
>>106858206Trust the science.
>>106858189>Some is going to get charged with a crime that he didn't commit all because the AI generated a face that resembled him from a shitty security cameraCan't wait
>>106858206I think if if they let the upscale reverse search images it would probably do OK with famous people's faces(rather than having to come up with it on the spot) but probably not with randoms
>>106858189> attach a blurred img > ask to AI to recreate a nigg president> flex on X to get ppl attention
>>106858287Even given how famous that photo of Obama is, it still fucked up the nose and mouth positions. Then again, if it's connected to the internet, why wouldn't it just reverse image search and return the actual photo, given the fact that it will always be better than an upscale.
>>106858189This is some uncanny void shit right here
>>106858189Now try it with a face that's not extremely ubiquitous in the training data due to being a literal POTUS
This only works so well because the AI has already trained on images of Obama specifically. It even got the freckles on his cheeks. There's no way it got those details just from upscaling the image on the left. I'd hesitate to even call this upscaling because you don't even need the image on the left when you could simply ask it for a picture of Obama and get the same result. All I see here is that it identified the picture on the left as Obama, but I think you could have reverse image searched that even before AI was trendy.
>>106858206The pigeon hole principle says otherwise
>>106858189That example is terrible because there is a million high res pictures of obama available to train an ai model on.
>>106858189now try using it on a non-famous image, retard
>>106858189>Shouldn't be possibleIt isn't; it's just searching for hires matches for the pixelated pic, not actually upscaling it. If you believe it could accurately fill detail over flat pixels, you're are fool.
>>106858373Fuck pigeons.
>we trained this ai on obama's face then gave it a blurry photo of obama and it basically shat out obama's faceyou don't say
Now do every episode of Kuroinu Kedakaki Seijo wa Hakudaku ni Somaru
>>106858208Do you just not moisturize? Or sweat, ever?
>>106858189Usecase for upscaling images of Obama?
what the fuck is this shit are people actually impressed by it?
>>106858189Well, it isn't, it's referencing all the hundreds of thousands of pictures of Obamer available on the internet. And he still looks weird.If you use a pixelated image of some literallywho, it's going to make shit up.Why is this freak losing his shit about this? Does he work at the company that makes it?
>>106858189That doesn't look anything like obama.
>>106858769it has to or you wouldnt have known who it was rerpresenting.
can this work on JAV?
>>106858189What a piece of trash, the turd on the right is yellow, on the left is red.
>>106858785The blurred pic looks more like obama than the AIshit rendering does.
>>106858769>That doesn't look anything like obama.Wym? It has the chimp face
>>106858189because it stole the solution from internet
>>106858787> try to upscale jav> japanese girl pussy gets 'upscaled' into Barack Obama's :I face
>>106858337You mean with a database with all DMV records? Yeah, (((they'll))) get to that.
>>106858189>Can upscale anything with AI>Upscales a nigger
>>106858189Can you use this on old games like shadow of the colossus?
thats a horrible job what are you talking aboutespecially since its fucking obama, the AI should know what he looks like much better than that
>>106859121Lel
>>106859458Just emulate them at high rez
>>106858206Easy fix. Just keep genning new images until one of them comes close to the original image. No one has to know about the mismatches.
>>106858189I guarantee there were a lot of Obama pictures in the training dataset. You won't get this sort of quality from a picture of a non-famous person.
>>106858189>the ai literally knows that image and thus creates obama>still does a shit job
>>106858224Is this a pov-ray scene? It actually looks quite prettyI'll consider myself humbled if it's image generation
>>106859539>>the ai literally knows that image and thus creates obama>>still does a shit jobIt's all recreational. it doesn't have a photo perfect memory.
The next president is going to be ObamAI isn't it? Republicans will try for a third Trump term, Democrats realized sassy black women are a recipe for disaster, and everyone is nostalgic for better times while having AI shoved into everything.
>>106859649sounds like a personal problem
>>106859656>The next president is going to be ObamAI isn't it?Absolutely, you fucking lummox. You can tell that to your sister/mom and everyone else at the trailer park. Spread the word.
>>106858199trust the training data
all that investment to.............upscale obama pics
That shit doesn’t look anything like the original photo, it’s not upscaling anything. He just generated some mid photo of Obama, what is the big fucking deal?
>>106859903>>106859483>>106859325>>106858769
>>106859928I suck at this.
>>106859903I love the fact that obama is becoming the new lena
>>106860110its fucking stupid because its a common face in any dataset
>>106860126And? so is lena
>>106858189Makes his face look like plastic.
>>106859928Obama is secretly a generic Australian man?
>>106860149Thats why he was hiding his birth certificate
>>106860130for one, lena is a specific image of a person who wasn't photographed anywhere nearly as commonly as obamaits not like you have a million different photos of her in every stock photo databaseand also I would make the same argument against using lena or any standard image nowbefore it wasn't such an issue
>>106859567Pure slop.
>>106859567
>>106860227>>106860260LGTM
I'll believe it when I see itand then I'll train my own model on video game textures
>>106858189>this shooden be possible!!! :N-word forehead:Those results are fucking terrible. What is this supposed to be advertising? Some subscription API for upscaling images?
>>106860149
>>106858769a gay man
>>106858189>this shouldn't be possibleholy shit shut up
>>106860300it's a standard sentence used by content creators to attract engagement through deception. making the mundane into the extraordinary. it's his job to trick people using psychology and flood the web with trash masquerading as something worthwhile.
>>106858189Retarded. Now show it a picture it wasn't trained on.
>>106858189how does a 25x25 pixel photo of the president manage to look exactly like the president after upscaling? it is a mystery
>>106858189>CRT vs LCD
Does it work on some random dude and not the most popular portrait of a US president?No.It's like compressing the entire Star Wars movies into the three bit sequences 101 and be mesmerized when the decoder (a 80GB data set) produces the entire Star Wars movies as output.
>>106858224>>106860227>>106860260These look like Myst / Riven stills
>>106861066High praise. Very inspired by Myst, esp. the realMyst remake. I was amazed Gemini could do the style so well, they've since nerfed it a bit.
>>106858189is this available somewhere? I kinda wanna feed it random noise
Finally I can depixel cunt flaps in JAV porn.This unlocks so much wall to wall pussy.
>>106858206Until they have a database of every single person's face
>>106858189Now feed it a random literal who's pixelated face.
>>106858189not even close enough to be a body double. white obama >>106859928 looks closer
>>106858224>>106860227>>106860260>>106861066>>106861235was going to say that looks like mystloved that game, it kept me sane
>>106858189Why do you need the image on the left to produce the image on the right?
>>106862975It's an image upscaler retard
>>106862990Yeah but if you can go from 32x32 pixels to 512x512 pixels then you can also go from 2x2 to 32x32 and you can just pick any 4 pixels you want out of thin air to get those 2x2 pixels so again why do you need the 32x32 image if you just want a portrait of some black guy?
>>106863044It's literally a reference to the infamous pic of previous AI upscalers: >>106859928
>>106858189Should've used a random guy. Not fucking barack obama that it has a million pictures of.
>>106858787All the AI "decensors" of JAV being spammed over the internet are fucking awful. I hate it.>>106859121kek
>>106860149>The drongos know, pack it in
>>106858189it only worked because it detected Obama and Obama photos are in training data. won't work for a random anon photo
>>106858189
>>106859928>left: Obama>right: Obamna
>>106859928this was more than 5 years agoAI isn't in it's infancy anymore, it should be better than this
>>106858189So what happens when it's a face that isn't a common public image and it hasn't been trained on?>>106858206People never think about how a task/problem is actually solved
AI sucks.
>>106858189Can it really be considered upscaling when there is so much generative work?Seriously. Upscaling never meant making things up
>>106858206That's precisely the goal tho
>>106858189Question: did the AI get prompted telling it that it's Obama?If so this is not impressive
>>106858189training on the test set
>>106858189>>106858620considering the amount of photos available of obama the result should have been better.It doesn't even maintain the same saturation level.
>>106858189>train lora on low res images>encode the image to latent, and upscale it using the lorai've been doing this for a long time already
>>106858189That's not an upscaler, that's a diffuser.
>>106858189Did they literally just hardcode Obama?
>>106858189It cheats :^)
>>106858373explain
