Knight Rider did one of those in the 80's. I was so disappointed when I got photoshop for the first time that I couldn't scan in a really blurry photo I'd taken as a kid and use "sharpen" to make it perfect.
Man, what three years has done in the field of image generation. From struggling with upscaling faces to creating entire images from text prompts. The pace of progress right now is astounding.
Something I noticed when using the software is that the de-noising features combined with upscaling an already high resolution photo that has some JPEGing can really help remove the more severe artifacts. Better than most denoisers I have tried over my 12+ years of photo editing.
I mean, it does kinda generate text poorly at times. Like how it once decided to generate "In order to stop this, you must stop this to stop this to stop this to stop this to stop this to stop this to stop this" and it just keeps going on and on. AI isn't the best thing ever, but hell it's surely better than me!
@@yeahboi4206 I'm not a boomer though, but I don't blame you for saying that, considering I'm the 11 year old who decided to login to my parents account.
there's a few "HD texture" projects using tools like this, although the results aren't always.... the best... it's a neat tool but a lot of it really should be left to actual human artists (ie majora's mask hd, those guys have been making great work) ...it's a pretty neat novelty though and kinda fun to just batch apply to textures of games
Shameless self-plug here, look for my Neural pack at moddb, it's under this same nickname I have. It's still wip and it's more of a mix of various AIs with my own touches on top, but I'm sure it won't disappoint you.
you don't understand what you watched if you think this is a good suggestion that will provide good results. you do not understand pixels or resolution. and how SMALL those images are.
I love watching old videos about Gigapixel and comparing the results people get with what's capable with the program now. It's gotten so much better it's ridiculous.
11:17 I swear I've seen this as a thumbnail on one of your channels somewhere haha This was a fascinating video, really cool that AI's able to do stuff like this, and makes unskilled people like myself feel even more worthless by the day
I am thankful for AI upscaling. As a layout designer, it helps a lot. It saves me some time dealing with dumb clients sending low resolution images. Following up on them just to send a high resolution version is so time consuming.
Some time ago, before Ai was that potent, I worked on image enhancing based on fractal decomposition. It worked stunningly on natural images, since most things have a quite a repeating pattern on them. It didn't work on: Faces, text etc. The AI seems to work similar. Well most non-linear recursion of a linear function (AI) ends up as a fractal...
3:02 It's very interesting you mention that downscaling first made it produce a better output! I would suspect the reason behind the better image quality when downscaling the image before feeding it to the AI would be an 'artifact' of how it was likely trained. To generate training data for a problem like this, they likely took high resolution images and downscaled them so they had an input and fixed output to compare the AI output against so that it could learn what it should be generating from an image like that. This probably acts like a form of data normalisation and the AI is sort of built to understand those higher contrast downscaled images rather than blurry old photos.
I'm kind of curious how much of a painting the pictures would turn into if you took an original sized photo: up scaled it, resized to default and repeated for lets say 10 cycles. Maybe a project for Twitter if you don't want another video on similar things?
This "2kliksphillip" guy needs to cool it, on constantly being terrified about computers and technology. Why is he bragging about his fears?? Is this a young person thing?? Is he aware that people have been using tools to be more effective/efficient for quite some time??
@Sam Williams Exactly, the tech is already here, just needs more training. Maybe add more recognizable features to help it understand what it's working on, but again that also will take more training. So in 1 or 2 years it'll be pretty insane, 3-4 years and it'll be crazy. Like numbers and letters could be features it should try to recognize. Then you could do the enhance of a license plate. Ofcourse it's not sure it would be 100% accurate each time. But we'd get up there.
In a few years it will probably be part of the next generation of JPEG/MPEG standard. We'll be using it invisibly everywhere as a way to compress media. Think smaller downloads for games with megatextures, livestreams being easier on your internet, fitting more songs on your phone etc.
I have talked to someone creating AI's and working with them and he said that most people overestimate what AI can do. He also said that he doesn't think AI's won't overtake the control. Most people just think AI's are like in the Movies which isn't the case.
They are really different things, Waifu2x is just an upscale algorithm whereas AI Gigapixel tries to figure out how the original image is supposed to look and adds details. This means that when it gets it right AI Gigapixel provides better results than Waifu2x but when it doesn't the results are very ugly, as this video shows in the case of human faces.
@@JannPoo I think Waifu2x does add some detail but not to the same extent, Gigapixel is way more aggressive in that sense. For example, here: imgsli.com/NTQ4MQ out of a pixelated mess, it managed to make well defined facial features, hair and clothes, it only failed completely with the thumb, which turned into a triangle. However, it doesn't add detail in the same way Gigapixel does, Waifu2x only attempts to turn those pixels into a shape that makes sense, but it won't mess with the texture, rather, it will smooth all the textures out if you set a very strong denoise, sometimes resulting in even less detail than in the original, because it is not trying to make a more detailed image, but a very clean image.
Well i was using this in same way you did for older games' textures including down scaling before upscaling in case of very bad simple textures to smooth them out. So for games its really good....unless they are realistoc in style then its probably as hit and miss as real life photos
I tried using gigapixel on HL2's textures and damn it makes such a difference. I had to switch to Waifu2x for the TF2 textures tho, but it helps so much because it gets completely rid of the noise produced by the VTFs, on top of sharpening the simple shapes and details.
Love the Topaz products, which include AI DeNoise and AI Sharpening. As with your experiments, these also benefit from some experimentation. But when it works it is so impressive that, as they say, you can never go back. Not to worry, though, about the need for experimentation with various images and textures: do you remember when digital audio first hit the street??? "Too clean, too sanitized," yada yada. Then they discovered that the analog-to-digital and digital-to-analog conversion chips and the analog amplification chips were actually unable to handle the new digital accuracy and needed upgrading. The rest, as they say, is history. Loved your video and love to see more of your work with Topaz. Mike
Interesting. I might be working on this stuff in the future. Im specialized in graphics which is super math heavy, and so thats put me in the ballpark when i switch to AI. Though graphics and image processing arent exactly the same.
I have an idea to take this experiment further. Using your techniques, taking a modern pictures in low resolution and then high resolution, using your AI upscale, upscale the low resolution and then compare to the modern version of the upscale to determin the in difference of the data. Seeing the data various forms may lead to new analysis on the actual transformation.
Amazing! I don’t know how long these images takes to process before getting the results back but imagine using some kind of low data mode on your phone/mail/whatever in areas with poor reception, you could send an image only a fraction of the size and let the receiver upscale it. That last set of image results of you on a boat blew my mind, the ~700x400 image is amazing
The CSI memes are still correct because A.I upscaling essentially only invents news information it wouldn't ever be able to retrive actual information out of a 2 pixels long car plate on a blurry image.
@@-Burb not exactly, he gave the reason to comparing A.I to CSI enhancements be wrong but didn't pointed it out, he even said that tech turned it in reality.
This is brand new and still somewhat impressive. In 10-15 years the capabilities are going to be insane. People were so obsessed with general AI that they missed what was really going to transform society. Niche learning algorithms like this.
I've been playing with this for the past couple months. I too found the magic of downscale then upscale for old blurry shots. Then you take the original into photoshop with it and pick and choose the parts you like. Its a great tool for artists but nowhere near replacing us. Edit: Try adding noise before upscaling. It helps!
CSI: Zoom in. Enhance that!
2kliks: Zoom out and shrink it, then enhance, and zoom in!
KingHalbatorix Why i upscale my shrunk image and not the original one
I recognise that face, it is the same guy as last week. And the week before!
The Demon-Murderer did hit again!
sometimes you gotta take 1 step back to go 2 steps forward
CSI enhance is still a meme, because you cant uncover detail that isnt there xd
Knight Rider did one of those in the 80's. I was so disappointed when I got photoshop for the first time that I couldn't scan in a really blurry photo I'd taken as a kid and use "sharpen" to make it perfect.
God, how I wish her face stayed blurred.
@@vablo7198
That won't work, you would have what seems to be a high res photo with very blurry faces
@@vablo7198
The technology isn't *that* advanced if you know how it works
@@vablo7198 no //
@@vablo7198
It really doesn't
@@vablo7198
There is no good solution, that's what I'm telling you
"Stupid AI"
AI will remember that
oh shit i would definitely don't want to be an enemy to AI fuck that.
“AI will store that in its database”
Rip telltale
Al, you are a magnificient suit of armor!
AI disliked that
Just imagine how a tree would feel looking at one of those monstrosities that we think look fine
Lmao😂
Chainsaw
I was thinking about this the whole video🤣
I am a tree, can confirm
oh my god hahahaha
Me: *watches on 144p due to shitty internet*
Phillip: It really adds clarity
Me: Ah yes
Bile it is still pretty good in 144p
@@nonsuspiciouscolour I wouldn't say
@@tmanwsoldtworld k
I can relate
Yip
The buildings look like PUBG when they dont load in fast enough
Reminds me of popping textures in games based on idTech 5/6
you should apologize to your Harddrive after this comment young man.
@@inertia_dagger ok
Kkl'
Wl
As
Z
"I'm landing play doh" is code for my game hasn't rendered, and I'm about to get stuck inside a staircase.
@Andrian looks like pubg when u look at it inside my pants jajaja
"which makes them look very creepy to us" *shows mark Zuckerberg*
couldn't be any more perfect
Saying "Eugh!" to Trump's face before the upscaling XD
@@GnRGaming How is that funny
@@mickel836 It's a joke on Trump that his face is worse than how it was upscaled.
@@hungershovel4357 yea, I know what you meant. But, how is it worse
@@mickel836He was either calling Trump ugly or made an editing mistake. If he's not careful he could start a flame war on political stuff
taking a moment of silence to appreciate the first person to milk a cow🙏🙏
This program already exists. Just Google it
Look up DeepCreamPy.
@@riccardoorlando2262 xDD
lol
Waifu2x, the best solution. And it’s free
Man, what three years has done in the field of image generation. From struggling with upscaling faces to creating entire images from text prompts. The pace of progress right now is astounding.
Stable Diffusion XL 1.0 says hi
"computers are becoming smarter"
my latest computer i got in 2019 can't run csgo without having a white line around every character model
Nice you have a result to beat those damn custom models
and now you are global elite
Is Bob Ross possessing this AI cause it likes trees so much?
no cabin, no bob
Bob Ross invented his own ai and he now decided to release it to the public.
Not important bob ross is dead
@@wcskeleton1388 Yea he invented his own ai and decided to release it to the public when technology had matched his chad genius upscaling.
no it likes to WOOD
uncanny valley: shows an image of Mark Zuckerberg.
Now there's a face that would benefit from more upscaling
You are damn right.
lmao
Zucki's wax statue looks more life like that zucki's face at the hearing...
and called creepy too uhhh haha
Some of the upscaled faces look like they were taken from the original max payne to me.
Is this the Payne residence?
@@dAnIeLmOnGe2000 Don't answer that, a rhetorical question
Lol
Something I noticed when using the software is that the de-noising features combined with upscaling an already high resolution photo that has some JPEGing can really help remove the more severe artifacts. Better than most denoisers I have tried over my 12+ years of photo editing.
1:55 looks almost real, but not quite..... Mark Zukerberg. I see what you did here. LoL.
He's creepy as hell!
What if it's actually the computers insulting how ugly humans are
That doesn't make sense because computers don't think.
@@MarioCardonaS BuT WhaT IF thAtS WhAt tHeY WanT yoU tO tHiNk?
😕Hmmmmmmmm
@@scrambled5948 Computers do not think. Think about it until you understand it.
@@MarioCardonaS You are the reason they are insulting us
"Stupid AI"
Be careful of what you say Philip
when the robot uprising starts, he is gonna die really early
I mean, it does kinda generate text poorly at times. Like how it once decided to generate "In order to stop this, you must stop this to stop this to stop this to stop this to stop this to stop this to stop this" and it just keeps going on and on. AI isn't the best thing ever, but hell it's surely better than me!
@@kygirlo528 ok boomer
@@yeahboi4206 I'm not a boomer though, but I don't blame you for saying that, considering I'm the 11 year old who decided to login to my parents account.
@@kygirlo528 reminds me of nier automata
Here's an idea, try running original Half-life or Counter Strike 1.6 textures through this program.
There's already a mod for half-life 1 being worked on with AI upscaling. (though water is broken on it).
there's a few "HD texture" projects using tools like this, although the results aren't always.... the best... it's a neat tool but a lot of it really should be left to actual human artists (ie majora's mask hd, those guys have been making great work)
...it's a pretty neat novelty though and kinda fun to just batch apply to textures of games
Shameless self-plug here, look for my Neural pack at moddb, it's under this same nickname I have.
It's still wip and it's more of a mix of various AIs with my own touches on top, but I'm sure it won't disappoint you.
They've already done this with FF7s textures.
Yup please
When AI saw your mom's face,
"I'm sorry Philip, I'm afraid I can't do that."
6:41 “looks hellfjsndifnalwbfj lot better”
hellofalot better
Just a quick stroke.
Maybe his voice was AI generated
helelolo lot better
GODDAMN BRITISH PEOPLE.
When he showed zuckerberg I couldn’t stop laughing
That is how he looked during the hearing... it was not upscaled at all... his wax museum face looks more human than him in reality.
Wait Zuckerberg is human ?!
I couldn't stop laughing when you said you couldn't stop laughing
@@aveareg laughinception
Actually I laughed when I read your comment
I am watching this on 240p and everything looks the same
Underrated comment😂🤣🤣
Try AI video upscaler
Wth yeah
I guess I see a difference because I'm watching in 360p. And that's the highest I can go since our internet connection can't take more than that. 😂
İ am watching it on 144 did you guys see pictures i think they are just colors
Do it with Minecraft(or any other pixel art game) textures.
you don't understand what you watched if you think this is a good suggestion that will provide good results. you do not understand pixels or resolution. and how SMALL those images are.
@@mailliw94 How about explaining it to him. Pointing out he didn't understand the video doesn't make him magically smarter
I don't think the point is to get good results, just to see what will happen/what the program would do with it
I didn’t say they would get better. I just wanted to see if something interesting happens.
William Detwiler I’m not saying that suddenly irl grass will pop up from grass textures.
I love watching old videos about Gigapixel and comparing the results people get with what's capable with the program now. It's gotten so much better it's ridiculous.
Everything: Rendered on supercomputer
Faces: Rendered on PS 1
Theres much better enhancement AI out now, like TecoGAN, featured on Two Minute Papers' channel
11:17 I swear I've seen this as a thumbnail on one of your channels somewhere haha
This was a fascinating video, really cool that AI's able to do stuff like this, and makes unskilled people like myself feel even more worthless by the day
Achieved with sorse
deja vu i've been in this place before
THE CSGO VIDEO
"It allows the ai to splurt its creative juices all over the photo in the most delicious of ways"
saucey as he zooms into his moms ass, what thr fuck
😳
Uh
iCum
AiJuice seems hot
This is actually quite terrifying but until i should actually be scared I will just say "Zoom in and enhance!"
efe
@@Maxmaxmaxmaxmaxmaax thank you for your input
Luca what’s up fellow luca
@@lucasalt6812 Oh hello there (:
Luca OH NO NOW MY PHOTO LOOKS BETTER!!!! RUUUUUUN!
Oh, how much we've advanced
I am thankful for AI upscaling. As a layout designer, it helps a lot. It saves me some time dealing with dumb clients sending low resolution images. Following up on them just to send a high resolution version is so time consuming.
Some time ago, before Ai was that potent, I worked on image enhancing based on fractal decomposition. It worked stunningly on natural images, since most things have a quite a repeating pattern on them. It didn't work on: Faces, text etc. The AI seems to work similar. Well most non-linear recursion of a linear function (AI) ends up as a fractal...
Talks about uncanny valley then shows a picture of zucc lmao
Now imagine this in game development in repurposing old texture assets.
This is actually already happening! I've seen people upscale the textures of Age of Empires 2 to create high-resolution mods a few months ago
Same thoughts
Nvidia just added image sharpening lol
@@WangleLine you mean official AoE 2 HD remake? :D
@@CZRaS No, I meant this! www.reddit.com/r/aoe2/comments/bsn7tt/playing_around_with_esrgan_on_aoe2_assets_an/
Damnn this video is already 4 years old. How good AI was 4 years ago is actually pretty insane
*Friend walks in and sees baby's eyes fullscreened on my monitor* 3:23
'Hmm. I better keep this blur.'
"CAN YOU ENHANCE THAT?" "Sure, but do you _really_ want that"
3:02 It's very interesting you mention that downscaling first made it produce a better output!
I would suspect the reason behind the better image quality when downscaling the image before feeding it to the AI would be an 'artifact' of how it was likely trained. To generate training data for a problem like this, they likely took high resolution images and downscaled them so they had an input and fixed output to compare the AI output against so that it could learn what it should be generating from an image like that. This probably acts like a form of data normalisation and the AI is sort of built to understand those higher contrast downscaled images rather than blurry old photos.
6:40 "Still looks a helulululot better than.. this"
was getting ready to timestamp this too lol thx
It's a tradition by now
Was looking if someone wrote this lmao
What’s the on going joke from?
@@hmorris from himself, he can't pronounce it correctly. Which is funny and nothing to worry about.
2:54 This is a really smart way to somehow censor an image while not censoring an image at all. Nice job, Phil :D
1:24
I love how he says “ugh” before AI upscaled lol
I'm kind of curious how much of a painting the pictures would turn into if you took an original sized photo: up scaled it, resized to default and repeated for lets say 10 cycles. Maybe a project for Twitter if you don't want another video on similar things?
It would be interesting to run the same low res image several times through it. The ending result will surely be nightmare fuel.
Just go to the end of the video
@@MaksKCS Ye but it wasn't run through several times in a row, he just decreased the res.
AI upscaling faces: *confusion*
AI upscaling trees: *happiness noises*
The last image straight from Max Payne 1 comics.
True Nightmare
1:19 i like how he says uuugh before trump is even upscaled
im so happy someone caught that
Fuck blumpf
Orange man Good.
This "2kliksphillip" guy needs to cool it, on constantly being terrified about computers and technology. Why is he bragging about his fears?? Is this a young person thing?? Is he aware that people have been using tools to be more effective/efficient for quite some time??
@@jamesmonroe3043 cry about it boomer
I nearly pissed myself at the ending lmao.
Such a good and well produced video though. Really cool seeing how the AI interprets every image.
I wonder how good this tech will be 10 or 20 years down the line.
Too bloody good
GMT VR, here we goooo!
@Sam Williams Exactly, the tech is already here, just needs more training. Maybe add more recognizable features to help it understand what it's working on, but again that also will take more training. So in 1 or 2 years it'll be pretty insane, 3-4 years and it'll be crazy. Like numbers and letters could be features it should try to recognize. Then you could do the enhance of a license plate. Ofcourse it's not sure it would be 100% accurate each time. But we'd get up there.
In a few years it will probably be part of the next generation of JPEG/MPEG standard. We'll be using it invisibly everywhere as a way to compress media. Think smaller downloads for games with megatextures, livestreams being easier on your internet, fitting more songs on your phone etc.
It's here in google's phones.
Me: **watches this video**
Also me: **puts minecraft through this program**
results please
ruclips.net/video/eC7f9tMslVE/видео.html they actually have an AI that upscales low resolution minecraft images
I have talked to someone creating AI's and working with them and he said that most people overestimate what AI can do. He also said that he doesn't think AI's won't overtake the control. Most people just think AI's are like in the Movies which isn't the case.
man, what a wild ride it's been.. I feel like no one expected it was headed where it was... :/
Finally we know how they generated the faces in Oblivion!
Waifu2x has been around for a while and there is a video version Video2x both are open source.
They are really different things, Waifu2x is just an upscale algorithm whereas AI Gigapixel tries to figure out how the original image is supposed to look and adds details. This means that when it gets it right AI Gigapixel provides better results than Waifu2x but when it doesn't the results are very ugly, as this video shows in the case of human faces.
Waifu doesn't generate any new detail like this program does, it just gives you cleaner atifact-free lines and edges.
@@JannPoo I think Waifu2x does add some detail but not to the same extent, Gigapixel is way more aggressive in that sense. For example, here: imgsli.com/NTQ4MQ out of a pixelated mess, it managed to make well defined facial features, hair and clothes, it only failed completely with the thumb, which turned into a triangle.
However, it doesn't add detail in the same way Gigapixel does, Waifu2x only attempts to turn those pixels into a shape that makes sense, but it won't mess with the texture, rather, it will smooth all the textures out if you set a very strong denoise, sometimes resulting in even less detail than in the original, because it is not trying to make a more detailed image, but a very clean image.
Waifu2X for drawings and cartoons.
Topaz AI for photos.
Well i was using this in same way you did for older games' textures including down scaling before upscaling in case of very bad simple textures to smooth them out. So for games its really good....unless they are realistoc in style then its probably as hit and miss as real life photos
Have you tried HL1 or maybe TF2?
I tried using gigapixel on HL2's textures and damn it makes such a difference.
I had to switch to Waifu2x for the TF2 textures tho, but it helps so much because it gets completely rid of the noise produced by the VTFs, on top of sharpening the simple shapes and details.
@@drachenpanzer5622 I bet it would make them sharp. Just like Wc3 since its so similar woth its textures
Love the Topaz products, which include AI DeNoise and AI Sharpening. As with your experiments, these also benefit from some experimentation. But when it works it is so impressive that, as they say, you can never go back. Not to worry, though, about the need for experimentation with various images and textures: do you remember when digital audio first hit the street??? "Too clean, too sanitized," yada yada. Then they discovered that the analog-to-digital and digital-to-analog conversion chips and the analog amplification chips were actually unable to handle the new digital accuracy and needed upgrading. The rest, as they say, is history. Loved your video and love to see more of your work with Topaz. Mike
"It enables the AI upscaling to splurt its creative juices in the most delicious of ways" i died
Your upscaled face at the very end of the video looks like something straight out of Amnesia.
To AI the faces are normal because that how they will make them after they take over.
Imagine using this program with old videogame textures...
HD textures here we go
Gta sa. Aw shit. Were we go again.
this is the kind of intellectual and funny content I was looking for.
Pure gem.
4:50 I don't know why I laughed at this, AI just made the building smooth.
nice
"There's such thing as uncanny valley"
**Shows Zuckerberg's face**
GG, Philip
6:42 when he said helleleldkdjndjwkcn i felt that
Ah, yes, the 5318008, one of the best numbers you can have. Other top tier ones being 69 and 420, of course
01134
Nice
11037
42 is the one true answer.
How about 7355608
The way you talk coupled with the music makes it feel like I’m watching Blue Peter, which is nice
This is a great comparison for where we were just 3 years ago with AI upscaling. Leaps ahead now.
5:24 Old school Lara Croft from Tomb Raider
That moment when you open up a video about AI but you see a dude pressing buttons on a calculator and tough you pressed the wrong video
Interesting. I might be working on this stuff in the future. Im specialized in graphics which is super math heavy, and so thats put me in the ballpark when i switch to AI. Though graphics and image processing arent exactly the same.
Very clever upscaling, it's pretty much asking the program to create and fill in more detail out of nothing and add it to the picture. Awesome.
philip saying "hella a lot", it gets me every time
I noticed when you said "urgh!" at Trump's face before it changed. Nice one.
Sir Joey of The White narwhals bacon epic for the win
"If this sounds scary, good, because it is" Bullshit, this is amazing
@man It's kind of not?
This was already 3 years ago?
I love the gag sound on Trump's face before the upscale
I have an idea to take this experiment further. Using your techniques, taking a modern pictures in low resolution and then high resolution, using your AI upscale, upscale the low resolution and then compare to the modern version of the upscale to determin the in difference of the data. Seeing the data various forms may lead to new analysis on the actual transformation.
I will never buy stock footage again. Just download the thumbnail, run it through this ai Programm and you’re good to go
4:35 Looks exactly like Google maps 3D buildings. Have they been using this?
They don't know...
That Thumbnail was very beautiful
Amazing! I don’t know how long these images takes to process before getting the results back but imagine using some kind of low data mode on your phone/mail/whatever in areas with poor reception, you could send an image only a fraction of the size and let the receiver upscale it. That last set of image results of you on a boat blew my mind, the ~700x400 image is amazing
The CSI memes are still correct because A.I upscaling essentially only invents news information it wouldn't ever be able to retrive actual information out of a 2 pixels long car plate on a blurry image.
Guilherme sampaio de oliveira
Yeah that’s exactly what he said
@@-Burb not exactly, he gave the reason to comparing A.I to CSI enhancements be wrong but didn't pointed it out, he even said that tech turned it in reality.
Switching back and forth between resolution feels like sitting at the optometrist and they go "One...or Two?"
The ai likes trees
It is giving us a message
"AI is here"
Wow, HAL, I didn't expect to see you in this video.
This is brand new and still somewhat impressive. In 10-15 years the capabilities are going to be insane.
People were so obsessed with general AI that they missed what was really going to transform society. Niche learning algorithms like this.
this was a year ago, and it already better
I've been playing with this for the past couple months. I too found the magic of downscale then upscale for old blurry shots. Then you take the original into photoshop with it and pick and choose the parts you like. Its a great tool for artists but nowhere near replacing us.
Edit: Try adding noise before upscaling. It helps!
You should try comparing results from Gigapixel against Photoshop's Preserve Details and Preserve Details 2.0.
AI upscalling makes the background look amazing, but it also reveals our inner demons
That cat's whiskers look like they're those beams of energy coming from a tesla coil
LMAOOOO THE AUTOMATIC CHAPTERS ON THIS VIDEO
"Only the lonely"
"I like to eat"
its beautifully subtle that you go eww on the picture of trump before the upscaling, unlike the rest where its after. clever.
The AI: *Hmm.. yes. The floor here is made out of floor.*
1:58 Lol, explains the uncanny valley and shows a real picture of Mark Zuckerberg.
laughed so hard
allright sir, I see you are a man of culture like myself 4:57 On other news gigapixel good pixel
A lot of times the results look like someone painted them in acrylic/oil, someone amazingly skilled
These face photos legit look like some simulacra jumpscares
Also the grass at 7:10 looks deep fried
11:10 who else immediately recognised that lens flare from the thumbnails?
yah
Try this on a low res csgo screenshot. (or maybe an old 1.6 photo)
Edit: I'd like to think I helped make the new 3kliksphillip video happen.
Csgo = 1.6... you mean cs right
ninty no... 1.6 specifically what people call the original 1999 game (the one before Condition Zero).
@@SacredDaturaa 2000* 1999 is the closed alpha.
“That is where digital or drawn people look almost real, but not quite right.”
Normal picture of Zuckerberg
And he’s a real person? Nah, he’s a lizard in a robotic body.
1:21 love how at every upscaled face you go ugh but trumps normal face already makes you do that