Become a member to get my videos early & other perks :p. Also, I made a free Discord server called AIHQ for anyone who is interested in anything AI! Come join: discord.gg/xxsXgX9adC
its not actually good that humans have these tools, because now art will lose what makes it meaningful. Also, artists will get paid less and eventually become useless.
Do you have or know of anything that is mind-blowing in audio AI? Like we hear a lot of things trying to replicate say Michael Jackson's voice in AI but it can't match tones yet. Hasn't been anything that you came across that has or can hear a sound and then try to replicate what the person sounds like?
I highly recommend Suno Ai is building a future where anyone can make great music. Whether you're a shower singer or a charting artist, we break barriers between you and the song you dream of making. No instrument needed, just imagination. From your mind to music.
Would be cool if in 5-10 years we can buy an upscaler box to route our old and new game consoles through to the effect demonstrated in this video. We could even choose between different aesthetics. Mario 1 can look like Steamboat Willie, play Star Fox with an origami Arwing and Donkey Kong with Muppets
will be 100% a thing in 5-10 years... refactoring all the old games (without producing much costs) - and reselling it again for +100$ just because of the new graphics... still i m very existed for that... imagine Ocarina of Time in the Unreal 6 engine - with new and more enemys, some new riddles and so on ...imagine playing doom2 (from 1994) with doom eternal graphics... and so on ...
Most of these companies are, I think, reselling what will be nearly-free commodities by the end of 2024. Upscale is basically free already and "reimagine" will be as well.
Absolutely, right my dear friend. What these companies are trying to sell is the most fundemental part of every AI generation workflow: hallucination in image to image. These are already being used by many. However these companies provide two things: GPU power and great UI to temper with all those settings we thinker with as tech savy individuals. Just like you, I predict that at the third or fourth quarter of this year, these will be freely and easily available for everyone. Video, 3D and audio included.
@@disketa25 if you're paying for local upscaling, you're being robbed. I have an extremely mid graphics (3050) card and use upscyl locally for free. It takes maybe 20-30 seconds to go from midjourney full rez image (~1200x1200) to 4x resolution.
So basically it's image to image stable diffusion with a very nice checkpoint and a Lora or two. Which you can use your consumer grade GPU to do locally. You can run stable diffusion on as little as 4 gigs of vram but obviously more is better.
That's what I thought as well. I've been able do img2img upscales in SD locally with similar results for about a year now. I don't know if any of these featured things have anything new to add, but I wouldn't know it from what I've seen here. It looks pretty typical for good SD models, and the UI kindof hints it's built on SD.
Holy cow. This thing is amazing. I stuck an old photo of me and some friends as children. It went from blurry and hard to make out, to looking perfect ... except the children don't appear to be us any longer lol. It's like someone came along 38 years later and replicated our photo, but with other children in place of us wearing our original clothes and with a way better photographer and camera.
As with other current AI I've tried, this is sometimes incredibly effective, but sometimes screws up, as when it converted a foot to a hand :) In portraits, the subject's appearance tends to be altered beyond recognition - but often creates a beautiful new image based on the contents of the original. Edit: I must mention that the Create is great fun. I asked for ‘Japanese woman, 30 years old, very long black hair, kimono’. It produced an excellent selection and responded immediately to edits, e.g. short hair, red hair etc. Using one of the results, it was then able to generate an incredibly good quality enlargement. Very realistic. It’s also very good at changing the ethnicity of the subject. Just tell it ‘Chinese’, ‘Indian ‘ or whatever, and Ms Chu becomes Ms Singh.
It does worse things than that. I seen it convert a bag of groceries into a dead baby. Rip. I saw it convert a picnic table spread with pumpkins, beef, and vegetables into a pile of 8 or 10 dead puppies. Just because there was a puppy a woman was holding in my picture. I think it was because it was Jack and Sally Skellington. But still, Krea is wild. Nsfw
What I find shocking is that the AI Midjourney and Leonardo generations that blew me away less than a year ago now looks garbage once its had an AI makeover with this tool, its weird how the mind overlooks the fine details when you are in awe of the capabilities of these tools.
@@maccagrabme What I find wild is that I can take a horrible ai picture that looks barely reminiscent of Taylor Swift, and I can put it into this. It knows Taylor Swift. Add the prompts. It gave me a super realistic Taylor swift, better than sdxl. Amazing. This program is mind-blowing. So many uses for it. It upgrades old photos. Not just ai art. Amazing.
I just played a bit with Krea, and I have to say: I'm blown away! I create surreal images with Bing, which I find far superior at this task than any other AI I've tried (and I've tried many). But using Krea as upscaler, with the same prompt and very conservative settings for "AI strength", really puts the icing on top. Not all the times, but very often. I'm in love!
better than gigapixel pro? I fix old photographs with it and it's like the exact photo was taken with a modern camera in whatever resolution you choose, it's like it knows all the details that should be there and doesn't change anything that it shouldn't.
I am VERY reserved towards these AI upscalers, because they essentially invent new content. There may be cases where that is totally fine, like messing with manga and lara croft, but upscaling portrait photos is very difficult for me, because the upscaled photo does not show the person on the photo, it shows an artist's impression. Before photography came along, all we had to capture a scene or the face of a person were the hands and imaginations of artists. Nobody knows what King Luis XIV. looked like. All we have is how artists portrayed him. Photography for the first time allowed us to capture the true, unfiltered image. Everybody knows what Queen Elisabeth II. looked like - for good or worse, in her state portrait or riding a horse wearing a scarf. With the advent of AI, we stand to lose that once more. Our children may never trust an image like we trusted ours.
Like suno for music, it requires some fine tuning to be conservative sometimes, so you've some idiotic or creative stuffs you never wrote in the lyrics by example, and that's also the case here, I guess. That's also the human dimension behavior when you order sometimes and the creator sometimes want show you something more open-minded than the thing you request, because AI like human tends to be creative in the way it has been made or educated (but yes, you can do repetitive stuffs like Chinese in an industry). Every AI I see is inventing/creating stuff but on some of them, you've only the tuning to be more or less creative. In that case, more conservative and not creative should be the way to go for the classic upscaling. --- and some disgression here, because your comment inspired me. Well, I'm happy I met my girlfriend today, because my feeling, about your children, is that on Internet, pretty soon, we'll have more and more difficulty to know if we talk to a real person. Including RUclips not removing/detecting AI stuffs on fake channels for scamming people... Like every tool, AI is a new way to do Good and Bad things. Hopefully, they'll make AI to autodetect and destroy that kind of scammy content... We live a revolutionary thing today, we created sometimes like a Big Smart Calculator back in the past to cheat, but we're losing a bit more of our humanity in this. Thus, I actually really enjoy the AI as old nerd, that's why I'm here discovering new stuffs on youtube. But maybe it will push people to meet and talk to stranger, and invite for a coffee in the street, concert, train station or another public place again (even if, the android will come in some years as well, as prototype today, then final products). We're closer to "Her" movie, before iRobot. :o) Except it will not kill us the way Terminator has made, it will kill us because hackers/scammers wanted to, manipulated by human themselves. AI we know are just kind babies wanting to help... but, that's true we only see the "good AI". How about fake AI made by scammers to scam efficiency people intentionally? MMh... But I'm still confident it will be moderated by the "Good ones" to avoid that kind of parasite over the Internet and the main services. Governements, at least here in Europe, are aware of the situation, demanding details about how works each AI and what references they have (but it's more for the copyrighted content, not really for security reasons).
@@Floxit Glad I could inspire you to think. Such a long reply deserves an equally well written answer. I'm not as optimistic in my outlook as you though. AI is, when all's said and done, a tool just like a hammer or a saw. You can use those to build a wonderful home for your familiy, or you can use them to hit someone over the head and dismember the body. Ultimately, it's down to humans how they use the tools they have. However, the concerning thing here ist the way humans have traditionally approached new tools in a naive, slaphappy way, only to discover later that once more their carefree use of them did in fact have consequences. One needs only look at leaded gasoline, plastic waste, nuclear power with it's still unsolved waste problem, or the deforestation of the Italian peninsula in the roman era. I also fear your confidence in our political leaders may be sadly disappointed. Political leaders tend to be old people, because it takes a long time to rise to power, and tend to be set in their ways and have a lot of trouble listening to expert advice, because they fear sharing that power. My prognosis is that AI will go largely unchecked by confused and powerless leaders, and that a lot of earth will be scorched before we learn to control and regulate it as it needs to be. One can only hope that we manage to do it before it does irreparable damage.
I love Krea's enhance fuction, but I prefer use stylar ,cuz it has more different styles to generate different images, hope you can make more tutorial about stylar, thank you so much !
Absolutely astonished by Crea ai's upscaling magic! 🌟 This isn't just upscaling; it's like giving a second life to every pixel. 🎨 Imagine the potential in art restoration, enhancing historical footage, or even personal photography! Who else is excited about the creative possibilities this opens up? 📸✨
And art creation right? Looking at that laura croft upscale I had imagined the ability to draw something myself though I'm not that much of an artist, if i could draw something so poorly but then feed it to an AI and upscale it...
So this was what the old CSI series or some movies truly wanted with their upscaling crime images or videos on the spot, but only with the wrong software/execution method lol.
Very transformative to regular images, but it could be useful if you want to upscale textures. I wonder if you could get it to make decent seamless edges via prompts
It's free only because it's in Open Beta. These AI things always use GPU power, so don't expect any of them be free. Unless there's a version that runs on your local PC and uses your own GPU. These web based interfaces don't, they use a GPU farm that costs money
AI does seem to have problems counting parts of an object wgich I think is down to how it works. I did many pictures of aeroplanes with generative or enhancing AI and it struggles with the propeller blades and symmetry. The funniest are where the left and right wings not only don't match but one is high the other is low!
that's the issue I ran into, it struggles with symmetry no questions, one eyebrow being lower than the other or rear wheel of the car several inches shorter than the front.
2:30 that particular noise pattern in the fur, I could simulate in Corel, 20 years ago. like, it's THAT algorithm... I recognize the pattern: it's the Whirpool distortion from Photopaint 11, from 2002. (itself inherited from PP 8, from 1999)
We're getting super close to where we're going to load an old AtariST game in an emulator, and run the game thru an AI upscaling program, pick a style, and with enough processing power - play that game in 3440x1440 fullscreen glory and no lag. I hope this happens and I live long enough to "play" with it. Of course a readily available movie upscaling would do wonders too, no need to buy re-editions of old movie flicks, now you just load your movie into the program overnight, and try it the next morning. Loving this !
Tried it with many different prompts and negative prompts, and without. Currently Photoshop superresolution neural filter does a better job if you just upscale 4+ times and then downscale again.
What do you mean, though? Krea isn't a traditional upscale. It adds new details that aren't in the original picture. Then you upscale the krea version in Photoshop or else Topez AI or something. The prompts in krea allow us to do all sorts of wild stuff.
I tried both AI editors. Both made me laugh. Especially the image enhancement one. It created a grotesque. People with 4-5 arms, Under shirt instead of pants, lush coat of Tibetan yak. The video generator created a series of images taken from iStock.
holy sh*t! the future of retro gaming is here! sure, it'll take some time before a tool can transform video game assets in real-time but the writing is absolutely on the wall now. exciting time to be alive! 👾
Wow. I'm an amateur 3D artist, using Blender, and whilst I usually aim for photorealism, I ran a couple of my more stylised abstract renders through krea. One it lost details that were important, and changed a masculine character to a more effeminate one but the other (an artistic interpretation of a part of Simon and Garfunkel's Sound of Silence), it added details that fit perfectly, but which I wouldn't have dreamed of adding. I may prefer the AI reimagining to my own original render, now.
Just found this video in my recommendations and it looks awesome. I have some early 3d ish pictures from the late 90s, early 2000s that aren't available in high res. And some very early b&w photos of my grandparents. So I may try this one someday. Thanks for showing us this. AI has surely come a long bit, even in music. Not 100% but close to it.
The cauliflower image is a sick burn because it's used by Dimension, a company hopelessly touting fractal upscaling since the 90s. Now they're being shown how it's really done 🤣
It'll be good when they start putting it into mod makers tools, such as Fallout 4 or Skyrim Creation Kit. AI generating Texture, and depth maps for games.
I’d like an AI application that can replace actors/actresses and voices so that movies and shows can be released with different versions to accommodate demand rather than some “token” character that doesn’t fit.
bro god bless you i found this video, this ki is that much insane. My pictures were that pixly and it makes them better then a real life person. This is insane
I would love to know, "Who owns the copyright of the A.I. generated image"? The A.I. software, or the software company or the user who manually texted the "text" promts with creative instructions to generate the image?
It upscales it by generating a whole new image yes. You can get it very close to what you want if you prompt correctly but it's not for every use case. If you want a normal ai upscaler I put a link in the desciption. Thanks for watching :)
Bro, because if you want to use the original face, you copy and paste it in there on Photoshop. Wtf. It's worth it. Krea looks more realistic than magnific ai even, even though magnific ai is often sharper and sometimes more sophisticated and technologically advanced perhaps. It also looks overly sharp, and it looks painted. Krea looks however you need it to for that specific case.
Thanks for this video! I had an out of focus picture of a bat face enhanced by krea: mind blowing. The picture is now clear and sharp. And yes, it does some strange things to some pictures. Some are good and some are aweful. But it's a lot better than the picture creator from Bing. That is: it works different. I can use my own pictures for it, which bing doesn't do.
I'm 567lbs and struggle with food and cannabis. Thank you for your AI video with AI voice. It helps reduce the voices in my head that tell me to harm my own genitals. Your video helps when I'm panic breathing and resort to punching myself.
@AIQuestOfficial Hi, can you recommend an AI program that can take a photograph and convert it into clean black and white line art like children's colouring book?
Hmm you can actually try using canva, if you go to edit image and change it to greyscale, then you can up the contrast/brightness. There are some AI apps that can turn photos into line art but thats not really like a colouring book. VanceAI says it does what you want, never used it though. I think Photodirector might also do it. If you just google “ai image to line art” you’ll get some results ;)
@@AIQuestOfficialThank you for your reply, I tried googling before I posted here, all I got was low quality noisy contrast adjustment type AI, something I could do in Photoshop for free by adjust brightness contrast. I would like to re-imagine the image as a 2d drawing. I will look into your suggestions.
I thought I might just add that the reason why the penguin's beak colour changes is because the original blurred image had the beak open and the enhanced image had it closed.
It will be interesting to see if AI can create graphical layers on top of classic games to upscale them and make them look incredible. Like a Skyrim graphical layer on top of Morrowind or Burnout from the Xbox era.
The beak closed and the feet are a bit buggered, but as long as you're not looking to fool money out of someone, the results look good. This video is 10 months old so I wonder if it's still free or have moved to charging by now. *Edit: There's a free tier but all tiers have limits clearly defined below them
We are reaching another tipping point in technology where the laws an our ability to differentiate between real world and not is becoming harder. Game engines are coming out with AI characters that can learn an understand they are not real an have real conversations, Which to me is a scary thing cause when does it become too much
Ai can be really Good at disturbing mind games without trying. I've seen something's I'm like you could wage war with this image. Truly break that man.
Big elections coming in 2024 in many countries. Not just USA. I'm predicting that like 2 days before election, someone will release DeepFakes of a candidate they want to lose. Like a video of Joe Biden saying/doing something very questionable. They will use the best DeepFakes they can (they have months to prepare), to hurt the elections. If there's only 1-2 days time, democracy will be affected by it. Even if people later learn it was fake, there will be a fast effect. It will be used JUST before voting day, so that there's no time to clear his reputation. I don't think the world is ready for 2024
Krea still needs a lot of improvement in how it handles certain things and pictures. For example, I have a picture of Jack Skellington with Sally and a boy. Sally is holding a puppy. They are sitting outside at a picnic table, eating pumpkins, vegetables, and beef. The krea image changes everything on the picnic table to piles and piles of dead puppies. I counted 8 dead puppies. With graphic details. Stuff that will haunt my nightmares it took a fork next to a plate and made it a tong, ripping a puppy's head off instead of a fork next to some beef and potatoes meal. I can't get over how nsfw and evil this thing's mind is. And how realistic the nsfw it creates without asking. I dont see a control to turn it off. 5 generations all contain the unwanted nsfw. Wtf. If I take puppy out of the prompt, it messes the puppy in her hand up. If I add puppy to the prompt, her actual puppy is okay, but the nsfw is there. Wtf. I had to give up on enhancing this image. Maybe magnifica would work perfectly. I haven't tried it. Too expensive.
@@CoolhandLukeSkywalkr What you're describing is simply to complicated for the AI to understand. You're never going to get what you want from an image gen with an image that complicated only using prompts. Your image is a _perfect_ use case for inpainting, and you'll be done in minutes, not hours.
@@CoolhandLukeSkywalkr Yes, you already do, inpainting is Stable Diffusion oriented word but just refers to selecting a part of an image and painting (prompting) something else in. In terms of workflow I mean more like, don't waste time on getting the entire image in one prompt, especially using PS it won't happen. Photoshop makes using AI easier through limiting features. In your example of Jack and Sally with puppies, you're probably right on that Krea is too automated and will consistently mess up the image. So the approach would probably be better and faster to drop the the original image into PS (assuming it's your editor/ AI of choice) and select and prompt each thing you want to do at a time. Or, if you were starting from nothing, to just get to a base image you could do that from. If there were no base image of Jack and Sally and puppies but, say, Krea generated your nightmare image from nothing, the nightmare image is still fine to work off of if the composition is correct, just pop it into PS and generative fill away the horrors one at a time. It will be _waaay_ faster than maybe getting a good generation, especially because unless the seed never changes the same prompt will yield different results anyway. I picked this workflow up on Civitai. It seems to be fairly common for more complex scenes, especially when you start getting into multiple people, or multiple subjects interacting. I'd definitely recommend browsing a few Stable Diffusion videos on controlnet inpainting, controlnet outpainting, and controlnet tile. SD can do everything Adobe's AI can do and much much more (depending on your hardware). My favorite thing about SD, is that you have _full_ control over how much the AI tries to change the image. The learning curve is fairly steep, but once you figure out a workflow hat fits your usage, SD can be equally as powerful, in different ways, as Photoshop. -and again, it's free.
is it stable between frames. can you provide a reference image to take as the secondary reference alongside with the primary image. ie to keep it stable. otherwise the image could be absolutely random.
While there are some quite astonishing Ai results, I feel that most of the showcase pictures took a long time of tinkering i.e. human optimization. I tried various versions of stable diffusion and the ammount of zombified pictures I have seen was just staggering. I also don't like the look of most of the "enhanced" images, because they all look like straight out of marketing photoshop and have nothing natural about them.
The showcase pictures yes, they must have gone through quite a few before picking those, but I've had some really good results myself. And yeah they do sometimes look a little bit "too good" but this has a lot of use cases nonetheless. Thanks for your thoughts!
How many of the images you uploaded to test were from the internet and thus may have been part of the AI training set? I wonder if that makes a difference in the output?
So basically, it does it wrong every time. Some parts are good, other parts wrong, and to fix what is wrong you need a good human artist. Almost useless now, but these are the first years, it will become better. It is LLM now, it needs more AGI.
Become a member to get my videos early & other perks :p. Also, I made a free Discord server called AIHQ for anyone who is interested in anything AI! Come join: discord.gg/xxsXgX9adC
Thats not a coliflower its a brocoli.
its not actually good that humans have these tools, because now art will lose what makes it meaningful. Also, artists will get paid less and eventually become useless.
Obviously the Mario → furry was just the new evolution in AI..
it clearly just accessed your subconscious,
and showed you what you wanted to see
Do you have or know of anything that is mind-blowing in audio AI?
Like we hear a lot of things trying to replicate say Michael Jackson's voice in AI but it can't match tones yet. Hasn't been anything that you came across that has or can hear a sound and then try to replicate what the person sounds like?
I highly recommend Suno Ai is building a future where anyone can make great music. Whether you're a shower singer or a charting artist, we break barriers between you and the song you dream of making. No instrument needed, just imagination. From your mind to music.
Would be cool if in 5-10 years we can buy an upscaler box to route our old and new game consoles through to the effect demonstrated in this video. We could even choose between different aesthetics. Mario 1 can look like Steamboat Willie, play Star Fox with an origami Arwing and Donkey Kong with Muppets
Omg yes to this!
Something similar might appear from Nvidia, GPU bound though, for this very thing..DLSS 5 maybe
I've thought before, even if they could do it with faces, uncanny valley might get pretty shallow
will be 100% a thing in 5-10 years... refactoring all the old games (without producing much costs) - and reselling it again for +100$ just because of the new graphics... still i m very existed for that... imagine Ocarina of Time in the Unreal 6 engine - with new and more enemys, some new riddles and so on ...imagine playing doom2 (from 1994) with doom eternal graphics... and so on ...
that's so creative
Why don't we have that Lara Croft?!
Because sweet baby inc exists
SWEET BABY RUIN GAMING 😭😭😭
Woke culture
She looks very close to Alicia Vikander, so we kinda do xD
Because we have Lara Croft at home.
Most of these companies are, I think, reselling what will be nearly-free commodities by the end of 2024. Upscale is basically free already and "reimagine" will be as well.
Here’s me hoping :)
If AI is moving at the pace it has been, what you said will happen before the end of february. 😅
Absolutely, right my dear friend. What these companies are trying to sell is the most fundemental part of every AI generation workflow: hallucination in image to image. These are already being used by many. However these companies provide two things: GPU power and great UI to temper with all those settings we thinker with as tech savy individuals. Just like you, I predict that at the third or fourth quarter of this year, these will be freely and easily available for everyone. Video, 3D and audio included.
It never was and never would be. I mean, even locally runned Topaz upscaling solutions cost you $100 (and a beefy GPU)...
@@disketa25 if you're paying for local upscaling, you're being robbed. I have an extremely mid graphics (3050) card and use upscyl locally for free. It takes maybe 20-30 seconds to go from midjourney full rez image (~1200x1200) to 4x resolution.
So basically it's image to image stable diffusion with a very nice checkpoint and a Lora or two. Which you can use your consumer grade GPU to do locally. You can run stable diffusion on as little as 4 gigs of vram but obviously more is better.
That's what I thought as well. I've been able do img2img upscales in SD locally with similar results for about a year now. I don't know if any of these featured things have anything new to add, but I wouldn't know it from what I've seen here. It looks pretty typical for good SD models, and the UI kindof hints it's built on SD.
Ngl, I clicked the video because of the thumbnail.
Thanks 😉
That's why it's called click bait)
I think it's worth noting that it's not an upscaler, as opposed to the title, but effectively a image based prompt
Yes you’re right of course, it can be used as an alternative to an upscaler though👍
Holy cow. This thing is amazing. I stuck an old photo of me and some friends as children. It went from blurry and hard to make out, to looking perfect ... except the children don't appear to be us any longer lol. It's like someone came along 38 years later and replicated our photo, but with other children in place of us wearing our original clothes and with a way better photographer and camera.
lol yeah that’s what it does, if you actually want to upscale that pic use real-esrgan on simmering like Upscayl
As with other current AI I've tried, this is sometimes incredibly effective, but sometimes screws up, as when it converted a foot to a hand :)
In portraits, the subject's appearance tends to be altered beyond recognition - but often creates a beautiful new image based on the contents of the original.
Edit: I must mention that the Create is great fun. I asked for ‘Japanese woman, 30 years old, very long black hair, kimono’. It produced an excellent selection and responded immediately to edits, e.g. short hair, red hair etc. Using one of the results, it was then able to generate an incredibly good quality enlargement. Very realistic. It’s also very good at changing the ethnicity of the subject. Just tell it ‘Chinese’, ‘Indian ‘ or whatever, and Ms Chu becomes Ms Singh.
It does worse things than that. I seen it convert a bag of groceries into a dead baby. Rip. I saw it convert a picnic table spread with pumpkins, beef, and vegetables into a pile of 8 or 10 dead puppies. Just because there was a puppy a woman was holding in my picture. I think it was because it was Jack and Sally Skellington. But still, Krea is wild. Nsfw
Definitely an interesting tool
What I find shocking is that the AI Midjourney and Leonardo generations that blew me away less than a year ago now looks garbage once its had an AI makeover with this tool, its weird how the mind overlooks the fine details when you are in awe of the capabilities of these tools.
@@maccagrabme What I find wild is that I can take a horrible ai picture that looks barely reminiscent of Taylor Swift, and I can put it into this. It knows Taylor Swift. Add the prompts. It gave me a super realistic Taylor swift, better than sdxl. Amazing. This program is mind-blowing. So many uses for it. It upgrades old photos. Not just ai art. Amazing.
@@maccagrabme This program also got sdxl James deal [without any lora] much closer to the actual James Dean. Amazing!
Looks like Magnific AI understands what we want.
Yes 😅
I just played a bit with Krea, and I have to say: I'm blown away! I create surreal images with Bing, which I find far superior at this task than any other AI I've tried (and I've tried many). But using Krea as upscaler, with the same prompt and very conservative settings for "AI strength", really puts the icing on top. Not all the times, but very often. I'm in love!
Awesome! Yeah I also use DALL-E 3 (bing) for a lot of things, different generators for different use cases. I love Krea too :)
better than gigapixel pro? I fix old photographs with it and it's like the exact photo was taken with a modern camera in whatever resolution you choose, it's like it knows all the details that should be there and doesn't change anything that it shouldn't.
I'll be honest here: I'd happily die of snu snu with that thumbnail version of Lara
🙈
I am VERY reserved towards these AI upscalers, because they essentially invent new content. There may be cases where that is totally fine, like messing with manga and lara croft, but upscaling portrait photos is very difficult for me, because the upscaled photo does not show the person on the photo, it shows an artist's impression. Before photography came along, all we had to capture a scene or the face of a person were the hands and imaginations of artists. Nobody knows what King Luis XIV. looked like. All we have is how artists portrayed him. Photography for the first time allowed us to capture the true, unfiltered image. Everybody knows what Queen Elisabeth II. looked like - for good or worse, in her state portrait or riding a horse wearing a scarf. With the advent of AI, we stand to lose that once more. Our children may never trust an image like we trusted ours.
Like suno for music, it requires some fine tuning to be conservative sometimes, so you've some idiotic or creative stuffs you never wrote in the lyrics by example, and that's also the case here, I guess. That's also the human dimension behavior when you order sometimes and the creator sometimes want show you something more open-minded than the thing you request, because AI like human tends to be creative in the way it has been made or educated (but yes, you can do repetitive stuffs like Chinese in an industry). Every AI I see is inventing/creating stuff but on some of them, you've only the tuning to be more or less creative. In that case, more conservative and not creative should be the way to go for the classic upscaling.
--- and some disgression here, because your comment inspired me.
Well, I'm happy I met my girlfriend today, because my feeling, about your children, is that on Internet, pretty soon, we'll have more and more difficulty to know if we talk to a real person. Including RUclips not removing/detecting AI stuffs on fake channels for scamming people... Like every tool, AI is a new way to do Good and Bad things. Hopefully, they'll make AI to autodetect and destroy that kind of scammy content... We live a revolutionary thing today, we created sometimes like a Big Smart Calculator back in the past to cheat, but we're losing a bit more of our humanity in this. Thus, I actually really enjoy the AI as old nerd, that's why I'm here discovering new stuffs on youtube. But maybe it will push people to meet and talk to stranger, and invite for a coffee in the street, concert, train station or another public place again (even if, the android will come in some years as well, as prototype today, then final products). We're closer to "Her" movie, before iRobot. :o) Except it will not kill us the way Terminator has made, it will kill us because hackers/scammers wanted to, manipulated by human themselves. AI we know are just kind babies wanting to help... but, that's true we only see the "good AI". How about fake AI made by scammers to scam efficiency people intentionally? MMh... But I'm still confident it will be moderated by the "Good ones" to avoid that kind of parasite over the Internet and the main services. Governements, at least here in Europe, are aware of the situation, demanding details about how works each AI and what references they have (but it's more for the copyrighted content, not really for security reasons).
@@Floxit Glad I could inspire you to think. Such a long reply deserves an equally well written answer.
I'm not as optimistic in my outlook as you though. AI is, when all's said and done, a tool just like a hammer or a saw. You can use those to build a wonderful home for your familiy, or you can use them to hit someone over the head and dismember the body. Ultimately, it's down to humans how they use the tools they have.
However, the concerning thing here ist the way humans have traditionally approached new tools in a naive, slaphappy way, only to discover later that once more their carefree use of them did in fact have consequences. One needs only look at leaded gasoline, plastic waste, nuclear power with it's still unsolved waste problem, or the deforestation of the Italian peninsula in the roman era.
I also fear your confidence in our political leaders may be sadly disappointed. Political leaders tend to be old people, because it takes a long time to rise to power, and tend to be set in their ways and have a lot of trouble listening to expert advice, because they fear sharing that power.
My prognosis is that AI will go largely unchecked by confused and powerless leaders, and that a lot of earth will be scorched before we learn to control and regulate it as it needs to be. One can only hope that we manage to do it before it does irreparable damage.
I love Krea's enhance fuction, but I prefer use
stylar ,cuz it has more different styles to generate different images, hope you can make more tutorial about stylar, thank you so much !
I’ll take a look thanks!
Absolutely astonished by Crea ai's upscaling magic! 🌟 This isn't just upscaling; it's like giving a second life to every pixel. 🎨 Imagine the potential in art restoration, enhancing historical footage, or even personal photography! Who else is excited about the creative possibilities this opens up? 📸✨
Exactly!
And art creation right? Looking at that laura croft upscale I had imagined the ability to draw something myself though I'm not that much of an artist, if i could draw something so poorly but then feed it to an AI and upscale it...
@@MrBrew4321 I have a video on exactly that! The Future is Here: REAL-TIME AI Art Drawing | FREE
ruclips.net/video/V7yUWTN6Jog/видео.html
Finally, the "zoom & enhance" memes can be a reality.
I know, crazy!
So this was what the old CSI series or some movies truly wanted with their upscaling crime images or videos on the spot, but only with the wrong software/execution method lol.
Exactly!
This will be wonderful for upscaling old pixelated game art especially to clip into thumbnails thank you. Subbed btw, sweet channel !
Awesome, thank you!
Very transformative to regular images, but it could be useful if you want to upscale textures. I wonder if you could get it to make decent seamless edges via prompts
Interesting idea!
Oh man that would be a game changer for creating textures. Imagine if you could just grab any low res image and generate a high res tiling one.
@@williamreinhard exactly my point but it seems the model needs to be trained on seamless edges, otherwise it generates mush
I now need this for my old VHS home videos
I have a video about free video upscaling, check it out :)
Very good tool, especially good for correcting logos, etc. Thank you very much for the video.
Definitely, glad you like it! Thanks for watching :)
This AI is amazing! It's completely free and you can create everything! (literally, it will do every prompt, no matter what you write)
Yep!
It's free only because it's in Open Beta. These AI things always use GPU power, so don't expect any of them be free. Unless there's a version that runs on your local PC and uses your own GPU. These web based interfaces don't, they use a GPU farm that costs money
@@shredd5705 I found a better one anyway. Perchance AI. It's truly amazing.
AI does seem to have problems counting parts of an object wgich I think is down to how it works. I did many pictures of aeroplanes with generative or enhancing AI and it struggles with the propeller blades and symmetry. The funniest are where the left and right wings not only don't match but one is high the other is low!
that's the issue I ran into, it struggles with symmetry no questions, one eyebrow being lower than the other or rear wheel of the car several inches shorter than the front.
AI and todays Tech is impressive and scary at the same time to me.
Definitely agree
This is the perfect Lara with a pretty face and triangular cheast, it's all the olds ever needed!))
Fair enough :)
ai is cringe
Agree to disagree😉
Why?
What's the difference between AI doing something and you using a computer program for the same task?
U are Cringe
People are cringe too
I cant wait for old games to one day run through live versions of this stuff, and play PS1 games completely upscaled.
Definitely!
2:30
that particular noise pattern in the fur, I could simulate in Corel, 20 years ago.
like, it's THAT algorithm... I recognize the pattern: it's the Whirpool distortion from Photopaint 11, from 2002. (itself inherited from PP 8, from 1999)
That’s insane
We're getting super close to where we're going to load an old AtariST game in an emulator, and run the game thru an AI upscaling program, pick a style, and with enough processing power - play that game in 3440x1440 fullscreen glory and no lag. I hope this happens and I live long enough to "play" with it. Of course a readily available movie upscaling would do wonders too, no need to buy re-editions of old movie flicks, now you just load your movie into the program overnight, and try it the next morning.
Loving this !
The video upscaling is already a thing, and it’s free! Check out the link in the pinned comment for my video on it. Also yes to the old games :D
Tried it with many different prompts and negative prompts, and without. Currently Photoshop superresolution neural filter does a better job if you just upscale 4+ times and then downscale again.
Good to know!
What do you mean, though? Krea isn't a traditional upscale. It adds new details that aren't in the original picture. Then you upscale the krea version in Photoshop or else Topez AI or something. The prompts in krea allow us to do all sorts of wild stuff.
I tried both AI editors. Both made me laugh. Especially the image enhancement one. It created a grotesque. People with 4-5 arms, Under shirt instead of pants, lush coat of Tibetan yak. The video generator created a series of images taken from iStock.
Yeah it can give some pretty funny results, I like the Tibetan Yak fur though
Best uncensored image generator ever
holy sh*t! the future of retro gaming is here! sure, it'll take some time before a tool can transform video game assets in real-time but the writing is absolutely on the wall now. exciting time to be alive! 👾
Wow. I'm an amateur 3D artist, using Blender, and whilst I usually aim for photorealism, I ran a couple of my more stylised abstract renders through krea. One it lost details that were important, and changed a masculine character to a more effeminate one but the other (an artistic interpretation of a part of Simon and Garfunkel's Sound of Silence), it added details that fit perfectly, but which I wouldn't have dreamed of adding. I may prefer the AI reimagining to my own original render, now.
Wow thanks for sharing! I’d like to see the before and after images if you want to share it in my discord :D
Just found this video in my recommendations and it looks awesome. I have some early 3d ish pictures from the late 90s, early 2000s that aren't available in high res. And some very early b&w photos of my grandparents. So I may try this one someday. Thanks for showing us this. AI has surely come a long bit, even in music. Not 100% but close to it.
Thank you! Yes i definitely agree. Krea might not work so well on the b&w photos but give it a go. I would suggest a regular Real ESRGAN for that
thank u sir, I will give it a try 🙂@@AIQuestOfficial
That Laura Croft one is not up scaling. That is straight up replacement.
Yeah that’s what it does, if you want free upscaling, real-esrgan is the best
@@AIQuestOfficial You just admitted it is a replacement.
It’s known as creative upscaling
@@AIQuestOfficial Sure.
The cauliflower image is a sick burn because it's used by Dimension, a company hopelessly touting fractal upscaling since the 90s. Now they're being shown how it's really done 🤣
Oh nice! :D
Amazing all this is available for free. Even the instant image generation is impressive. Super fun to play around with.
It really is!
It's no longer free :(
Isso que eu estava procurando esses dias, pegar um compilado de cutscene de jogos antigos e fazer esse upscaling com IA pra deixar mais realista
Sim
That image of Lara Croft is the best render of her EVER! They're redoing her look again and this has my vote 100%
Good 👍
I think the Lara Croft sample is cheated, it's a existing skin from a database overlay.
It'll be good when they start putting it into mod makers tools, such as Fallout 4 or Skyrim Creation Kit. AI generating Texture, and depth maps for games.
That's going to be amazing
@@AIQuestOfficial It hasn't happened yet, but it will eventually happen. Even Photoshop has AI now...
This raises some questions about the nature of our universe for me, if such complexity can be derived from such simplicity, where does that leave us?
Interesting question
It's not that simple though, it has been trained on insane amount of data, pictures taken from the around the world.
Nude mods for films coming soon lads.
It would be cool to have a temporally stable real-time upscaling for old video games
This was an awesome video!! And now I’m going to binge the channel
Yay! Thank you!
I’d like an AI application that can replace actors/actresses and voices so that movies and shows can be released with different versions to accommodate demand rather than some “token” character that doesn’t fit.
Facts, I wish I could replace all white actors with a more diverse cast instead
Imagine if that existed when ps1 started.That Laura render was in point
Yep :D
bro god bless you i found this video, this ki is that much insane. My pictures were that pixly and it makes them better then a real life person. This is insane
Glad you like it!
the painting touch is really nice
Yess
It's like giving someone a sketchy picture and asking him to reimagine what it looks like. That's how AI works...
Exactly
I would love to know, "Who owns the copyright of the A.I. generated image"? The A.I. software, or the software company or the user who manually texted the "text" promts with creative instructions to generate the image?
I believe Krea owns the art but you have a commercial license to sell if you have the pro plan
It completely replaces the model's face.
If the AI strength is too low, nothing happens. If it is high, the face disappears.
It upscales it by generating a whole new image yes. You can get it very close to what you want if you prompt correctly but it's not for every use case. If you want a normal ai upscaler I put a link in the desciption. Thanks for watching :)
Bro, because if you want to use the original face, you copy and paste it in there on Photoshop. Wtf. It's worth it. Krea looks more realistic than magnific ai even, even though magnific ai is often sharper and sometimes more sophisticated and technologically advanced perhaps. It also looks overly sharp, and it looks painted. Krea looks however you need it to for that specific case.
Thanks for the recommendations, is there any real-time image upscaler?
(Sth like anime4k but for anime images rather than videos?)
1:54 purrfect!
wow this is so cool keep it up love your videos
Thank you! Will do!
@@AIQuestOfficial I hope to see you grow big in the future im sure you will
Like it or not AI is the future. Big companies will use AI to make perfectly targeted movies that are completely AI generated. A hit every time.
Yep, I doubt we’ll really recognise the world in 20 years
Thanks for this video! I had an out of focus picture of a bat face enhanced by krea: mind blowing. The picture is now clear and sharp. And yes, it does some strange things to some pictures. Some are good and some are aweful. But it's a lot better than the picture creator from Bing. That is: it works different. I can use my own pictures for it, which bing doesn't do.
Awesome!
when using it, it makes a prompt for that image, so you could probably use the prompt in a image generation tool like midjourney
Indeed!
"Creepy...dont look"...and then i bursted out laughing, that is some funny shit.😂😂😂
Can you do another tutorial, but only do laura croft as examples?
Hmm maybe, not sure how interesting that would be for everyone 😅
Penguins have feathers not fur or polyester fibers.
Fair enough
The best improvement is the cake, where the raspberries turned into strawberries ... 🤣🤣😂
Yesss
I am going to use it to sharpen my digitally zoomed photos take with my iPhone standard model.
Real-esrgan is better for this, Upscayl is a free program!
Heyyy, I could probably something like this to improve some of the graphics and image quality of the some of the games I'm working on.
Perfect use case!
I'm 567lbs and struggle with food and cannabis. Thank you for your AI video with AI voice. It helps reduce the voices in my head that tell me to harm my own genitals. Your video helps when I'm panic breathing and resort to punching myself.
Now imagine it capable of doing this realtime on any game. or even going through a games files an upscaling the textures and stuff that way.
Exactly!
It seems not too far from Midjourney's upscales. I don't see a difference other than it's easier to use and more expensive. But to each their own.
This is quite different from MJ. Krea doesn’t just add more pixels, it reimagines the whole image and enhances it 👍
I think this technique would be amazing for video game remasters!
Will i finally look good when i upscale a pic of myself?
Definitely lol
It will change your face I guess😂
@AIQuestOfficial Hi, can you recommend an AI program that can take a photograph and convert it into clean black and white line art like children's colouring book?
Hmm you can actually try using canva, if you go to edit image and change it to greyscale, then you can up the contrast/brightness. There are some AI apps that can turn photos into line art but thats not really like a colouring book. VanceAI says it does what you want, never used it though. I think Photodirector might also do it. If you just google “ai image to line art” you’ll get some results ;)
@@AIQuestOfficialThank you for your reply, I tried googling before I posted here, all I got was low quality noisy contrast adjustment type AI, something I could do in Photoshop for free by adjust brightness contrast. I would like to re-imagine the image as a 2d drawing. I will look into your suggestions.
Humans: Ai is there anything you can't learn?
AI: No.
Humans: okay draw hands.
*AI shaped hole in the door.
Haha
ooooo, we are getting close to the CSI "enhance... ENHANCE... ENHANCE!!!!" tech
I know!
this generative AI is really mind blowing. I know in probably a year you can probably turn any picture or video into nudes accurately using AI
I thought I might just add that the reason why the penguin's beak colour changes is because the original blurred image had the beak open and the enhanced image had it closed.
Ah yes indeed
It will be interesting to see if AI can create graphical layers on top of classic games to upscale them and make them look incredible. Like a Skyrim graphical layer on top of Morrowind or Burnout from the Xbox era.
Oh hell yes Morrowind
"That’s not upscaling, that’s just magic." 🎉😮
what amazing years we leave in i love Ai
I know right :D
Love how the AI fused Musk's hands together.
Glad you were entertained :)
Holy crap the man.. the myth, the legend...
You turned poor little Mario into Stay Puft Marshmallow Man! 😶🌫️ #AI will conquer the Earth!😆
😅
Thanks for this, Krea is amazing to mess around with.
No problem! Yeah its pretty great, thanks for watching/commenting :D
1:53 "Which I will link in description"
Narrator: "He didn't"
Oh oops, it’s the top link if you google Real-esrgan Replicate, or you can use the free local program called Upscayl
I hope it evolves a lot and goes for video upscale, making several low-quality videos in 4K
Yes!
The beak closed and the feet are a bit buggered, but as long as you're not looking to fool money out of someone, the results look good. This video is 10 months old so I wonder if it's still free or have moved to charging by now.
*Edit: There's a free tier but all tiers have limits clearly defined below them
We are reaching another tipping point in technology where the laws an our ability to differentiate between real world and not is becoming harder. Game engines are coming out with AI characters that can learn an understand they are not real an have real conversations, Which to me is a scary thing cause when does it become too much
Ai can be really Good at disturbing mind games without trying. I've seen something's I'm like you could wage war with this image. Truly break that man.
Big elections coming in 2024 in many countries. Not just USA. I'm predicting that like 2 days before election, someone will release DeepFakes of a candidate they want to lose. Like a video of Joe Biden saying/doing something very questionable. They will use the best DeepFakes they can (they have months to prepare), to hurt the elections. If there's only 1-2 days time, democracy will be affected by it. Even if people later learn it was fake, there will be a fast effect. It will be used JUST before voting day, so that there's no time to clear his reputation. I don't think the world is ready for 2024
Haha, that frog was blended with a chameleon.
Lol yeah
it's not free.I'm supposed to buy some tokens.
You should get free daily generations🧐
6:28. Well. If I have to wait in a line or pay 30 bucks for this I might as well not waste me time on this good idea in principle.
Awesome content keep it up!
Thanks so much!
Amazing video!! Already done some really cool upscales with it!
And it did mess up Elon's hands, if you look carefully :)
Thanks! Nice 👍
Krea still needs a lot of improvement in how it handles certain things and pictures. For example, I have a picture of Jack Skellington with Sally and a boy. Sally is holding a puppy. They are sitting outside at a picnic table, eating pumpkins, vegetables, and beef. The krea image changes everything on the picnic table to piles and piles of dead puppies. I counted 8 dead puppies. With graphic details. Stuff that will haunt my nightmares it took a fork next to a plate and made it a tong, ripping a puppy's head off instead of a fork next to some beef and potatoes meal. I can't get over how nsfw and evil this thing's mind is. And how realistic the nsfw it creates without asking. I dont see a control to turn it off. 5 generations all contain the unwanted nsfw. Wtf. If I take puppy out of the prompt, it messes the puppy in her hand up. If I add puppy to the prompt, her actual puppy is okay, but the nsfw is there. Wtf. I had to give up on enhancing this image. Maybe magnifica would work perfectly. I haven't tried it. Too expensive.
Oh damn yeah thats horrible😬
@@CoolhandLukeSkywalkr What you're describing is simply to complicated for the AI to understand. You're never going to get what you want from an image gen with an image that complicated only using prompts. Your image is a _perfect_ use case for inpainting, and you'll be done in minutes, not hours.
@@CoolhandLukeSkywalkr Yes, you already do, inpainting is Stable Diffusion oriented word but just refers to selecting a part of an image and painting (prompting) something else in.
In terms of workflow I mean more like, don't waste time on getting the entire image in one prompt, especially using PS it won't happen. Photoshop makes using AI easier through limiting features.
In your example of Jack and Sally with puppies, you're probably right on that Krea is too automated and will consistently mess up the image. So the approach would probably be better and faster to drop the the original image into PS (assuming it's your editor/ AI of choice) and select and prompt each thing you want to do at a time. Or, if you were starting from nothing, to just get to a base image you could do that from.
If there were no base image of Jack and Sally and puppies but, say, Krea generated your nightmare image from nothing, the nightmare image is still fine to work off of if the composition is correct, just pop it into PS and generative fill away the horrors one at a time. It will be _waaay_ faster than maybe getting a good generation, especially because unless the seed never changes the same prompt will yield different results anyway.
I picked this workflow up on Civitai. It seems to be fairly common for more complex scenes, especially when you start getting into multiple people, or multiple subjects interacting.
I'd definitely recommend browsing a few Stable Diffusion videos on controlnet inpainting, controlnet outpainting, and controlnet tile. SD can do everything Adobe's AI can do and much much more (depending on your hardware). My favorite thing about SD, is that you have _full_ control over how much the AI tries to change the image. The learning curve is fairly steep, but once you figure out a workflow hat fits your usage, SD can be equally as powerful, in different ways, as Photoshop.
-and again, it's free.
is it stable between frames. can you provide a reference image to take as the secondary reference alongside with the primary image. ie to keep it stable. otherwise the image could be absolutely random.
While there are some quite astonishing Ai results, I feel that most of the showcase pictures took a long time of tinkering i.e. human optimization. I tried various versions of stable diffusion and the ammount of zombified pictures I have seen was just staggering. I also don't like the look of most of the "enhanced" images, because they all look like straight out of marketing photoshop and have nothing natural about them.
The showcase pictures yes, they must have gone through quite a few before picking those, but I've had some really good results myself. And yeah they do sometimes look a little bit "too good" but this has a lot of use cases nonetheless. Thanks for your thoughts!
Incredible
Thought so too😊
finally ai is actually useful
:D
@@AIQuestOfficial more fun than useful
Played first Tombraider 1996, Lara looks still cute and athletic.
How many of the images you uploaded to test were from the internet and thus may have been part of the AI training set? I wonder if that makes a difference in the output?
I don’t think so, but give it a go you’ll see it works great with your own photos too👍
magical how the frog now has...*checks biology notes* ... 4 nose holes
Lol
So basically, it does it wrong every time. Some parts are good, other parts wrong, and to fix what is wrong you need a good human artist. Almost useless now, but these are the first years, it will become better. It is LLM now, it needs more AGI.
Its crazy good rn idk what you're on about. Also it does need "more AGI" lmao