AI video just got better! Updates you need to know about

5.33k views5148 WordsCopy TextShare
AI Search
Minimax image to video, Pika 1.5 full testing & review #ainews #ai #aivideo Thanks to our sponsor K...
Video Transcript:
there have been several notable updates in the world of AI video so one of the best video generators out there minia Max has finally released an imageo video feature so in this video I'm going to test this out I'm going to test out a series of very tricky images to show you what it can and cannot do plus I'll compare these results with cling which is another close competitor finally we also have paika 1. 5 which was actually released around 10 days ago at the time of this Rec recording but their servers were stuck when they released it and I couldn't actually generate anything until this week anyways in today's video I'm also going to do a full review on Pika 1. 5 I'm going to compare it with the two leading video generators out there cling and Minimax to show you what it can and cannot do let's Jump Right In first of all for Minimax it's pretty straightforward to use I'll link to this in the description below where you can sign up for a free account after you have signed up you can just just enter in a prompt and then click create now I've already gone over their text video feature in a previous video so see this if you haven't already but recently they've added this new feature which lets you upload an image as the start frame so if you click on this it allows you to upload an image so first I'm going to select this image which I created with flux and then for this prompt section you can either leave it empty or to give it more guidance we are going to type in a terrified girl running away from a T-Rex and then all you need to do is Click create all right and here is what we got you can see this generation is not great so the girl is transforming into a woman and her face her body is completely different plus the T-Rex's arm kind of fell off but it regrew a new arm and then it's not really chasing the girl it's just walking off in the distance so not a very impressive Generation by the way I'm going to do the same thing with clings so this is a close competitor this is also one of the best video generators out there they already have an image to video feature so if you log in and then click on this tab here is where you can upload an image so again I'm going to upload the same photo and then for the prompt I'm going to add in the same prompt and then we will click generate and this is what we get for cing you can see actually King's generation is a lot better the T t- Rex is chasing the girl the girl remains very consistent she doesn't turn into a woman and then the T-Rex also remains fairly consistent across the entire video and then here are the two side by side I would say in this case the clear winner is cling all right next up let's upload this meme image and see what that gives us I'm not going to enter in any prompt for this example so it can do whatever it wants with this image let's click create and see what that gives us all right here's what we got for Minx and oh my goodness she is pissed this guy is in big trouble and here's what we got for cling what's going on here I don't know what's going on here can someone explain this to me anyways here are the two videos side by side let me know in the comments which one you think is better all right next up I'm going to upload this anime image of two Kung Fu Masters fighting and now this is a tricky image because first of all it's testing if it can generate anime and second it's testing if it can generate a fight scene which not a lot of video generators can do well right now and then for the prompt I'm going to type in anime style a fight between two Kung Fu Masters and then let's click create and here is what we got for Minimax you can see it can pull off an anime style and they are kind of fighting however the hands and fingers just deform quite a lot this is a clear flaw I don't think this is a usable video and here's what we got for cing this is actually pretty cool I mean everything remains consistent however they're not really fighting I did not specify there to be fire in the prompt but that's what it generated it's kind of cheating here because it's not making them fight it's just adding some fire effects to the scene anyways here are both of them side by side and again I just want to emphasize that that fight scenes are very hard and pretty much none of the top video generators can actually pull off a very coherent consistent fight scene all right next up I'm going to test out this meme again I'm going to leave the prompt empty and see what it gives us let's click generate and here is what we got this is a perfect video Everything remains very consistent but again this is quite an easy picture I mean the girl is just standing there looking at the camera there's no high action movements in the video so this should be a pretty easy generation for it to pull off and here's what we got for cling and you know this is pretty cool it's actually rare to see this effect where first the camera focuses on the girl in the foreground and then it kind of Zooms in to focus on the background this is a very cool effect and again everything remains very consistent I would have to say I prefer clings generation a bit better it's just slightly more detailed and sharper anyways here are the two videos side by side let me know in the comments which one you prefer all right next up I'm going to upload this image which I created with ideogram this is an awesome AI image generator and the prompt is just a young woman doing a live stream so let me actually paste The Prompt in here as well and then click generate all right here's what we got from Minimax you can see at the beginning there they actually added a pretty interesting effect where it zoomed in a bit and you often see this Zoom effect on YouTube when you have an influencer talking at the camera so that's a pretty interesting addition now notice her face isn't really consistent especially her eyelids this doesn't look great and then her hands and fingers are also not too good so if you look closely they do kind of warp into these weird deformations over time and it's just not consistent and here is what we got from cing you can see King's generation is a bit sharper plus her face is a bit more consistent her eyelids actually look consistent like I don't see any weird deformations on her face so that's pretty good her hands and fingers however are still flawed you can see these deformations over time and then here are the two videos side by side in this instance I would have to give the point to cling but let me know what you think all right let's try some crazier examples I'm going to upload this image of a Japanese woman being attacked by a zombie which I generated on ideogram and then for the prompt let's paste in a young woman with a terrified expression she is being attacked by a zombie from behind and let's see what this gives us all right here's what we got from minia Max oh my gosh she is terrified this is quite a terrifying scene and for the most part I mean everything remains kind of consistent except for her jacket there and there still is some hallucinations on her hand but other than that I mean this makes for a pretty decent scene in a horror movie and this is why I love image to video you can first generate an image as the start frame and there are a ton of tools to give you a lot of control like flux and stable diffusion and control net and lowas for consistent characters so this gives you 100% control of the start frame of the video and you can create some pretty crazy things from this all right and here's what we got from cing oh my goodness she's trying to like shake the zombie off her and then the zombie just kind of disappears you can see there's a lot more warping going on in this video it's not as consistent as Minx so I don't think this generation is particularly usable anyways here are the two side by side in this case for this High action scene I would have to give the point to Minx and you know what it's kind of fun generating videos of a young woman attacked by a zombie so let's generate another one I'm going to upload this image which was also generated by ideogram and then click create all right here's what we got and this is perfect it's hard to notice any flaws with this and again this can be a scene straight from a horror movie this is so good all right and here's what we got from clling by the way notice that I'm running cing 1.
0 here cuz I ran out of credits to do 1. 5 anyways this one it's a bit slow motion so cling is kind of cheating here I mean slow motion is pretty easy to generates you can see overall everything remains consistent she doesn't look as scared as the video from Minx though anyways here are the two videos side by side again for this example I would have to give the points to Minx it's better at handling these types of scenes all right my final test let's see if it can just get an anime character to talk because earlier the example of the anime Kung Fu Masters fighting it's very hard for it to generate a fight scene so let's make it easier I'm going to upload this photo of an anime girl with a really nice pair of let's focus on the video guys guys let's focus on the video okay so for the prompt it's really simple I'm just going to type in anime girl talking and see what that gives us all right so here's what I got from Minimax you can see it's not great especially her mouth her mouth looks really weird I guess it kind of got the anime eyes correct but still not perfect and then her hands and fingers also warp over time so overall this is not a very good or usable generation and here's what we got from cing actually cling does this slight better but she's not really talking I would like to see her you know open her mouth a bit more and she's not like moving her hands around as much as Minx so you can't actually see her hands and fingers in other words cing is kind of cheating here it's making it easier for itself it's not exposing any hands and fingers it's not making her talk so you would expect this video to be more consistent anyways here are the two generations side by side let me know in the comments what you think all right so that sums up my test of mini Max's New Image to video feature you can see for some of the time it's better than cling for example for the scenes with a woman being attacked by a zombie but for other instances cing handles it better so for example for this scene of a girl being chased by a T-Rex King's generation was clearly better than Minimax so I guess the verdict of this comparison is that there's no clear winner and unfortunately if you want to generate something it's best to try out both platforms and see which result is better anyways next let's move on to pabs we have a new AI video generator in town and it's called Pika 1. 5 it was actually released last week but their servers were stuck and I couldn't actually generate anything until this week anyways in today's video I'm going to do a full review on paika 1.
5 I'm going to compare it with two other leading video generators out there cling and minax so you can get a sense of what it can and cannot not do thanks to Catalyst for sponsoring this video catalyst is a super powerful AI tool to generate scripts and storyboards it makes it easy for filmmakers advertisers and content creators to turn scripts into vibrant storyboards in seconds so for example if you don't have a script you can simply type in an idea and it would generate a script for you so for example let's try a love story between Jack and Jill and you can within seconds it gives us a full script that looks like this which we can turn into a storyboard now you can edit each one of these rows and you can also edit the prompt further there are various settings such as the aspect ratio and also the art style you can go for a sketch style or cinematic cartoon pixel art or animation it also generates consistent characters across your storyboard so here you actually need to choose the face of your character and you can see there are a ton of different options you can choose from and you can see within seconds it's able to to generate a full storyboard with the images and with consistent characters you can also add a new character including your own custom character once you've created a new character for example you can say Elon enters the scene and you can see now it generates a scene with our new character of Elon you can also upload your own image to use as a reference frame so this gives you unparalleled customizability and you can customize the image of each card you you can change the angle the distance the location and even customize the poses of all the characters easily save your script into various formats plus when you're done you can also easily present it with this presentation mode join many creators already using Catalyst click on the link in the description below for a 7-Day free trial so to access paika all you have to do is go to Pika doart which I'll link to in the description below and then once you're in you can sign up for a free account and once you do sign up you do get some free credits to start with now Pika has actually been around for a while this is one of the pioneers of AI video and it existed way before even Sora was announced now of course back then AI video was pretty bad all we could really do at the time was simple zooming and panning and we couldn't really generate High action scenes like we get with current video generators but finally paa has gotten an upgrade and this latest version 1. 5 they claim it has much better quality and prompt following and consistency so I'm going to put this to the test I'm going to test its limits on a series of very tricky prompts but before we do that there's another feature they released and it's pretty fun to play with it's called Peak effects and here's how it works so let me just get rid of this prompt first for Pak effect you need to upload an image I'm going to upload this meme and then if we click on Pika effect there are a few options we can choose from we can either inflate it melt it explode it squish it crush it or turn it into cake so let's try let's try inflate it and then press generate and see what [Music] happens all right let's try another example I'm going to upload this image of a girl being chased by a T-Rex and then instead of inflate it let's explode this and see what happens so here are some more examples here's a crush it [Music] example here's a squish it example you can see it's kind of turning it into PL now to be honest here's the thing I mean these paika effects are pretty neat but I can't really think of any use case for this other than it's fun to play with so at least for me this feature seems like something I would play with for like one or two days and then never use again but let me know in the comments what you think of this and if you actually have a use case for this anyways next let's actually test out its video generation capabilities I'm going to test it on a series of pretty hard prompts of various styles to show you what it's good at and what are its limitations so the first prompt is aerial drone view of an Alpine mountain range at Sunset and it's pretty simple and straightforward to use you just need to type in your prompt here and then make sure you choose paa 1. 5 which is the updated model with the better quality according to them and then there's also this settings button over here which lets you set some further options you can also enter in a negative prompt these are all the things you don't want to include in our video so for example I'm going to enter cartoon animation 2D low quality and then you can also set the aspect ratio we are going to leave it at 16 to9 so once that's done let's click generate all right so here's what we got you can see the quality isn't great it's kind of blurry it's not high resolution plus it's not really a drone video it's a video of a drone so I guess it's taking my prompt way too literally I was looking for a video that's taken by a drone and it looks like the propellers of the Drne aren't actually spinning so this doesn't look too impressive by the way here's the same prompt with cling and Minimax and you can see both of them just look way more detailed and more realistic compared to Pika all right next one I'm going to try is an astronaut riding a unicorn in the desert again quite a tricky prompt there's a lot of elements involved it has probably never seen this in its training data so let's see if it can generate this well again I'm going to set the negative prompt the same and then click generate and here is what we got again not too impressive I mean the astronaut looks very realistic the American flag on his arm looks realistic as well the Unicorn looks great however you know the main problem is the Unicorn isn't really walking it's just kind of floating you don't see any stepping motions from the Unicorn so again this kind of looks like the AI video technology we've had from last year it doesn't seem to be able to handle higher action movements and then for your reference here is also the generations from Minimax and cing and you can see both of them can actually get the Unicorn to walk and I don't know it just seems like the movements are a lot more fluid and higher action for Minimax and cling all right next prompt this is even trickier a group of Pomeranian puppies learning to become chefs this is Trick because it's not just one puppy but a group of puppies and it's not just any dog it has to be a Pomeranian plus they're learning to become chefs it's probably never seen this in its training data let's see what it can come up with and here's what we got this is actually not bad these Pomeranians are all wearing Chef hats and they're all looking around curiously these are really cute by the way they do look like Pomeranians and you know they are learning to become chefs although they're just standing there and not doing anything I would like to see see them actually cook or move some food around but overall it's still not bad and for your comparison here is the same prompt but with Minimax and with clling and again you can see Minimax and clling they have higher action these puppies are actively working with the food and learning to become chefs whereas for Pika they're just standing there and being cute but overall I mean P's generation is not bad for this example let me know in the comments what you think let's now see if we can do this dis Disney Pixar style so the prompt is a princess wearing a beautiful glittery white dress running away from a massive dragon with glowing red eyes Disney Pixar Animation style and then for the negative prompt I'm going to remove all of these animation keywords because that's actually what we want to include so let me click generate and here is what we got so first of all the princess is kind of wearing a glittery white dress give it some points for that there is a massive dragon with red eyes the eyes aren't really glowing and then she's not running away she's standing still plus it doesn't seem like she has feet and she doesn't really have a face which is even scarier than the dragon and like I said she's just standing still she's not running away so zero points for that I mean the dragon is kind of Disney Pixar style but the princess we can't really see her face not too great of a gener ation and for your reference here's the same prompt but with Minimax and cling and you can see Minimax handles this Disney Pixar style very well for cing not so well and for both of them the princess is not running away from the dragon and so I actually have not encountered a video generator that can actually get this prompt 100% correct none of the princesses actually run away from the dragon all right here's another one the text subscribe to my channel made of vibrant colorful smoke then here again let me add the keywords Caron animation 2D low quality so here I'm testing its ability to generate text in the video Let's click generate all right here's what I got well it kind of pulled it off I mean subscribe to my channel is the correct text but in my prompt I specified for this text to be made of vibrant colorful smoke so it's not really made of the smoke it's just overlaying text on this vibrant smoke and then here's the same prompt but with Mini Max and cling so you can see actually all three of them have some flaws none of them are perfect but let me know in the comments which one you prefer it seems like paa is kind of the laziest here in that it's just over laying text on the video it's not actually integrating the text with the smoke all right next prompt now I'm testing if it can generate anime style and this is actually something that Pika does very well so the prompt is a girl wearing a kimono walking in the streets of Koto anime style and then for the negative prompt I'm going to get rid of these 2D keywords because that's actually what we want all right let's click generate and here is what we got you can see it actually nailed the face of this anime girl this looks exactly like anime she is wearing a kimono however her walking is really strange it looks like she's walking sideways like a crab or she's either floating along the street in any case the motion looks very abnormal so that's a clear flaw here but in terms of actually generating anime style videos and characters piga does quite well as you can see here and then here is the same prompt with Minimax and cing so in this example I think cling actually did the best here Minimax for some reason was unable to get this anime style that I was going for and then Pika it's actually good at generating anime characters but the motion is just weird all right let's try even harder prompts so here the prompt is a woman who is very sad and distressed her eyes are red and teary her facial expression conveys sadness and emotional pain for the negative prompt let me me add in cartoon animation 2D so I want this to be as realistic as possible let's click generate and here's what we got this is not bad actually she does look oh my God never mind did you see the eyes at the end there that is creepy as helmet that's going to be nightmare feel for me tonight anyways she does kind of look sad and distressed her eyes are kind of red and te but but she doesn't really convey enough sadness and emotional pain and what I mean by that is here are the generations from minax and cling and you can see both of them just look a hell of a lot sadder I mean you can clearly see the intensity of the woman's emotions in both Minimax and clings Generations but for Pika she doesn't really look sad enough all right next one is even trickier point of view shot of a soldier running through a war torn city rifle in hand the camera moves quickly as explosions occur nearby throwing up debris The View dips and swings as the soldier takes cover and fires back at the enemies let's click generate so this is a really hard prompt let's see if it can pull this off and compared to Pika version one this is not bad all right version one can only do simple zooming and panning it can't even get people to run here it's at least getting the soldier to run it is kind of a point of view shot however you know some people are not running normally you can see this dude on the right here I don't know what he's doing and then it's not really a high action scene it's not throwing up debris The View should dip and swing the soldier should take cover and fire back at the enemies I'm not seeing any of this here in comparison here's the generation from Minimax and cing and you can see both of them just handle High action scenes much better I mean you can see from this example Pika is clearly still a step behind all right next one is even more challenging horror film a swarm of zombies attacking people at a Metro Station Shaky camera let's click generate and see what we get all right and I'm actually very impressed by this generation I mean at least Pika version 1 cannot pull this off but here we do have a swarm of zombies they all have long hair for some reason you can see some flaws like over here their body is kind of warp in shape and it's not really consistent and then there's this dude here who is also warping over time I mean everything is still not very consistent if you look closely there's a lot of warping and deformations going on so not a particularly great generation now here's the same prompt with Minimax and cing and you can see for cing it's amazing clling was able to generate a shaky camera which I specified in the prompt and this looks very much like a horror film with zombies attacking people now for all three of them there are inconsistencies the zombies warp in shape over time and so none of them are perfect and that is because this is a particularly hard prompt whenever you generate multiple characters or multiple objects in a video that's where you start to see more hallucinations and errors but anyways those are the three vide side by side let me know in the comments which one you prefer all right next one a massive Evil Panda looming across the city destroying buildings terrified people run away in all directions High action let's click generate and here's what I got the panda actually looks very nice this does look like an evil Panda it is destroying the city kind of there's some weird things going on over here but overall this is actually quite an impressive generation and then here are the two generations from Minimax and cing again in terms of high action I think clling Nails it but you know this generation from Pika is actually not bad okay so that's sumbs up my video on minia Max's New Image to video feature as well as Pika 1.
5 as you can see for Pika 1.
Copyright © 2024. Made with ♥ in London by YTScribe.com