The Best AI Image Upscaler Just Leveled Up 16x!

Video Statistics and Information

Video
Captions Word Cloud
Reddit Comments
Captions
hey everyone today we're taking a look at a pretty big update to what is currently my favorite AI image upscaler and I know that's a pretty weird thing to say like how exciting can it actually be it's an upscaler right it just makes things bigger but I think when you see it your head will start spinning into a lot of creative uses we'll take a look at some controversial AI art news coming out of one of the world's biggest card games some 1.0 Hol deck news it's not quite here yet but we can see it just on the horizon and it is pretty exciting and we'll round out with some Insider tips and tricks on creating consistent characters in scenario which I covered in the last video okay let's dive in so magnific who I have covered on the channel in the past have released an update to their upscaler allowing for times 16 upscales or up to 10,000x 10,000 pixels it's a lot uh to give you an idea of what that looks like fauxfur created this video kind of this sci-fi-esque image of you know everything being sucked into a black hole uh as we zoom in in here you can kind of see like we've got these land masses waves that are kind of crashing in as we pull out we see some clouds and atmosphere but here's where things get crazy is now as we start pulling back we see the actual size of the entire image and I mean that's massive it's it's really really big still going that yeah that's really crazy it's really cool image too another cool and creative example from Martin neong uh this is a sculpt that he put together in ps5s dreams um and as you can see he's alternating back and forth between the resolution of the original uh image and what magnific has done to it I mean that is pretty insane how closely it ends up zooming in and retaining and creating details briefly the way it works is you will upload an image to magnific obviously from there you can choose to two Exit 4 exit 8 exit or 16 exit now a quick note on that 16x is that you are still limited by magnific uh 10,000x 10,000 pixel cap though that is pretty huge from there you have a number of different presets that you can choose from anything from standard to science fiction horror for this we'll be going with portrait soft underneath that is a prompt box in which you can reuse the same prompt that you used before uh for example this is a mid Journey V6 image I could just take that same prompt and bring it in but you will note that you can actually wait that prompt as well uh by using a parentheses say in the example here it would actually work well beautiful green eyes and then weighted at like 1.3 1.42 even um and then close parenthesis from there we move down to the slider section I probably should have mentioned this earlier if you're stressing out about this VIP only thing um don't worry about it this feature is supposed to be live later today for everyone and trust me I am not a VIP um anyhow the first slider is creativity and this is where I think magnific really gets its spank by turning the creativity up you're basically giving magnific the ability to hallucinate over your base image so it will create additional details sometimes those details are super awesome sometimes they're a little bit weird for now we're not going to go too crazy with it though I will turn that up in just a little bit uh the second slider is HDR or high dynamic range resemblance is the slider in which you can control how much magnific sticks to your original image so if you turn it up magnific will try to adhere as much as possible to your original image though possibly at the cost of some blotchy looks or kind of a dirtier overall look uh whereas turning it down gives magnific permission to kind of get a little fuzzy with the details fractality is the new slider and basically what that does by lowering the fratila but you are prone to less glitches whereas if you start cranking the fractiline permission to start making up details which can work in your favor or can get super weird uh there is an example of a rose here so this isn't too crazy but um as we can see if we look at the center of the Rose here there aren't a lot of water beads but after we take it through a super upscale um we end up with this where you know you can kind of see additional details on these water beads very much like a Georgia o'keef painting doesn't it so going back to our model here uh I didn't go too crazy with prompting or the sliders on this one um but you can see that there is definitely a difference after running it through a creative upscale um the hair texture I think looks a lot better not quite as soft um the eyes definitely pop a lot more and the skin tone I think looks a lot better as well um but this is actually the thing that kind of will blow you away so it's always kind of hard to wrap your head around exactly how big a upscale is so this is our magnific upscale obviously and um on another layer that is our mid Journey upscale so yeah it is pretty significant so at 100% this is where we are on our magnific upscale so that is uh pretty remarkable is there someone oh that's kind of cool actually in her reflection you can actually see a photographer or at least someone presumably taking that photo that's pretty wild I did notice that there was kind of one little background artifact here um not that huge a deal you could easily just clone or paint that out so just to give you an idea of things that you can do by merging up scaling and creativity within magnific uh I ended up taking a sort of a reference from an Italian Renaissance painter named pinio I believe his name is uh sort of like this style of imagery bringing it into mid Journey using it as an image reference and then adding in prompts for kind of a fantasy look and we ended up with this which is pretty cool and this is the non upscaled mid Journey version uh and then running it through magnific uh creative upscale with an upscale factor of eight um just if you look at the difference uh just between here and here you see how much more it actually ends up cleaning up the image it's really when you zoom in that you see exactly how much detail uh magnific is capable of putting in I mean that's pretty insane uh just for reference this is the non upscaled mid-journey image in that same area you know punched in as well now it does change things you'll note that like in this we have this you know fire and kind of a blurry painting going on here whereas on the magnific side it actually filled in the painting which is pretty impressive uh and then created I guess kind of like this opulent um golden plant sculpture it seems very appropriate for the time I guess but it really is interesting just to kind of zoom in on certain areas just to see what magnific decided to do I mean some of it's weird there's no question there but um I mean a lot of it is really cool as well now I don't want to overly sell this as like an absolute fix for everything for example I took a mid Journey image of Time Square uh from from sort of a bird's eye view and ran it through magnific and yeah I mean it looks uh you know it did its job but it's not going to do something like zooming like super in and turning all of these people into real people they're still going to be sort of Blobby people not quite as bad as like maybe the initial mid-journey output I mean granted this is zoomed in like 500% so uh I don't know of anything that's going to handle a crowd scene like this and be able to generate characters you know throughout this whole scene but I will not that it does some really nice stuff to like Street textures so this is the before mid Journey image and the after magnific image but in terms of those Blobby people uh magnific is a pretty good way of sort of taking your older mid Journey images and giving them a Fresh coat of paint as Julie W design did here uh by taking this older it looks like maybe mid Journey 51 uh image and then running that through magnific and you can see um it definitely does improve across the board now I do want to point out because I always get yelled at in the comments for this one magnific is not free and actually the pricing plan is a bit on the high side uh starting at $39 per month and going up to $299 per month and listen I know that's a lot my hope is at some point some Enterprise customers will end up joining on and more users will join on and hopefully that will allow the price to come down or perhaps even a lower tier moving on to AI AR Kur fluffle news Wizards of the Coast the owners of the card game magic the Gathering a game that I never played because well I suck at regular cards I can't handle spells in cards as well they ended up taking a pretty hard stance on AI generated imagery stating our internal guidelines remain the same with regard to artificial intelligence tools we require that artists writers and creatives contributing to Magic the card Gathering I think it is TCG the card Gathering sure uh to refrain from using AI generative tools to create final magic products well except fans were very quick to notice that AI generated assets were used in a social media ad campaign for magic of the Gathering taking a quick look I mean to my eye it definitely looks more along the lines of adobe's generative fill than it does say mid journey and I do think it's important to note that the cards themselves do have traditionally created art now Wizards of the Coast initially denied the accusation which probably wasn't a good look because it didn't take long before somebody took this image and pointed out details like this now what's sort of funny about that to me is the fact that you actually still do have accurate numbers on this side meaning that whoever was the artist on this probably did take some time doing some additional touch-up work Wizards of the Coast did then issue a statement essentially pinning this whole thing on a vendor that they were using but it is clear to me that whatever artist was working for this vendor um definitely did put in some time and work on this I mean we we do have correct text here uh the plasma ball sure that looks you know fairly Firefly esque but you know choices had to be made in terms of placing these cards um it does look like some Shadow was also added in there as well this isn't a case where you know someone just prompted four magic cards sitting on a magical shelf and ended up with this image I guess the irony in the whole thing is the fact that having this anti- AI art policy has ultimately ended up getting you know some artist working for a vendor in trouble or you know possibly fired so we fired a human so that the AI wouldn't replace them I don't know if the person got fired I certainly hope that the person did not get fired in holc news meta have released audio to photo real embodiments which basically takes audio and turns them into photo realistic animated uh character avatars like 10 years now I've I mean I've LED I've LED movies features yes weren't you in that feature last year you really like the main supporting I mean not Le but you were supporting to the main that was yeah yeah yeah yeah that was that was me so why are you that has definitely come ways from those like Sims AI avatars that Zuck was trying to push on us like a year or two back additionally over at CES right now uh we have the Holo box which uh you know apparently is a box that will generate a hologram of you I mean it is hard to tell what the overall quality of considering that this is obviously a 2d video video um it does look like it's just front-facing there's that camera that that guy just pointed to there so um I don't know if it's doing a full 360 but yeah I mean it's kind of cool you you've ever wanted to talk to yourself standing in an empty refrigerator your time is coming and I'm not going to get too crazy with CES stuff I mean obviously every Tech media Outlet in the world is there right now covering it um but I did want to point out that uh Apple is releasing the Apple Vision Pro that is going to be be in stores in February to demo uh and additionally Sony has released their XR uh headset as well utilizing the Snapdragon processor it's just an interesting Flex on their part considering that they already have the psvr so I don't know it definitely looks like VR is going to be a pretty massive push uh towards the Hol deck coming this year finally rounding out last week I did a whole Deep dive tutorial on how to create consistent AI characters in scenario scenario actually reached out we got to chatting via email and they actually had a couple of tips and tricks uh and one thing that I got wrong uh that I definitely wanted to share with you guys so obviously you'll probably want to have watched that first video uh it is linked down below so you know you can watch that and then come back and check this part out so starting off I guess with my kind of mistake um I had initially said that you could take one of your character's full body turnarounds and then just kind of shrink them down into the bounding box and uh that it you know the white on the side probably wouldn't matter as it turns out that is actually not the case scenario did point out that sort of the white background in the bounding box does have an effect on your overall model and what they would recommend doing is either knocking out the background or finding a similar color tone for what you are trying to relatively achieve with your model additionally a number of you were having some issues in terms of creating like sort of colorfully garmented characters uh the solution to that is actually just to take the template model and literally just paint over over them uh and apparently you really don't have to do it that well um and then if you prompt correctly you should be able to get a character with that attribute so trying that out really quickly and very terribly painting some blue jeans on our template character uh bringing him back over to scenario and using him as an image reference and then using the model cell shaded icon stylizer along with the prompt handsome Surfer blue jeans tan skin blonde hair and then uh waiting the front view back view and side view prompts with three parentheses open three parentheses closed uh and then running that and that got us this which is you know pretty accurate he might not be as tan as I had necessarily expected but um you know obviously bare feet there uh no shirt now there are other Generations in this series uh in which he was more tan uh and you know in this one he's wearing a shirt we've got a shirtless bro here no shoes on and lastly uh anime St Rogers who apparently is wearing uh shoes blue shoes in this one blue denim shoes on with denim all right one thing that I did forget to mention in the initial video as well is that if you find a character that you kind of are okay with like we're going to take anime Steve Rogers here you can always download him bring him in as a reference image add in prompts and then you know generate again from there and again I want to point out it is still AI uh it does take a little bit to get to like we did not end up with an Aloha shirt in any of these until the last one where we got almost an Aloha shirt so it still will take some work in terms of generating to get the character that you are looking for but once you actually go through this process you then have the character scenario did have a number of other tips and tricks for me that I will share with you and kind of probably more of a dedicated video on that note that is it for me today I thank you so much for watching my name is [Music] Tim
Info
Channel: Theoretically Media
Views: 24,343
Rating: undefined out of 5
Keywords:
Id: unqRqIDyHuI
Channel Id: undefined
Length: 15min 24sec (924 seconds)
Published: Tue Jan 09 2024
Related Videos
Note
Please note that this website is currently a work in progress! Lots of interesting data and statistics to come.