AI Video Leveled Up Again! Gen-2's Latest Update is Remarkable!
ฝัง
- เผยแพร่เมื่อ 28 ก.ย. 2024
- Today we're taking a look at Runway ML's Gen-2 update, which allows you to use any image and turn it into a video! We'll begin by looking at an incredible video featuring the return of Walter White, followed by a number of other amazing Gen-2 examples created by some very talented artist.
We'll take another look at Pika.Art outputs vs Gen-2, but also interestingly look to see what happens when you actually blend them together!
I'll show you the best way to achieve realistic faces in Gen-2 using Roop, and then close out with a short segment on where all this might be headed for the future of personalized entertainment, and the current SAG strike.
Links:
Pika vs Gen-2: • AI Video Showdown: Gen...
AI South Park: • This AI Generated Sout...
Roop: replicate.com/...
AFFILATE LINKS:
Camera: amzn.to/3yXMDY2
Microphone: amzn.to/3K1jIZm
Audio Interface: amzn.to/3lDX9kf
Mac Keyboard: amzn.to/3XOeJR0
Coffee: amzn.to/3JZuBeq
Follow Me on Twitter: / theomediaai
-------------------------------------------------
Thanks for watching Theoretically Media! I cover a large range of topics here on the Creative AI space, Technology, Tutorials, and Reviews! Please enjoy your time here, and subscribe!
the coherency and consistency is getting a lot better very quickly in these clips… we’re not far from having something that’s actually useable…. feels like we just need one more jump to have a “midjourney 5” moment with this stuff 💪
I also wonder what things will look like when Midjourney itself jumps into the game. The Motion feature has been in development for a bit now, and if there's one thing I've learned about MJ it's that they may run silent for a hot minute, but when they drop something, it is usually stunning.
Requires a lot of trial and error. Most results are bad and complete nonsense BUT can you imagine how useful this will be once the output results are a lot more accurate.
I can see now why they released that "unlimited" option. Too rich for my blood, but I'd be generating like a madman if I had it!
Its possible in 2040...
Ironically, the strikes might be accelerating the process. Big studios are suddenly entering the research field and with their enormous resources I can only imagine the acceleration in progress they're going to achieve
I do feel the same
I used to be scared of AI, but this makes it more exciting and is giving me profitable ideas.
Excellent to hear! Nothing to fear, it’s just an extension of creativity!
Glad to see this. I was just talking about how people could create cutscenes or movies, using picpik and an wi image generator to make your own motion capture video with pikpic.
Hahaha I had no Idea Bond was drinking like that!
He's been totally hammered for the last 70 years!!
The bond head movement looked to me like a conversation reaction as if re was about to respond with something he didn't full accept. "Yeah that's life for you" reaction 😊
Ha! I still think he's drunk...well, actually it could be both!
3:47 That's a door opening on the street she's on, not a wall of death! Haha
Thanks for doing so many different variations!
Great video as always!
It’s a DOOR OF DEATH!! Haha, I love AI insanity. I know it will eventually become as rare as extra limbs in Midjourney (remember when it was rare to get the correct number of limbs?), but a part of me will miss the wackiness when it’s gone!
As always , excellent content Tim! Thank you!
Thank you! I was pretty happy with this video! Came out pretty solid!
@@TheoreticallyMedia Correction. All of them are consistently solid! Hats off to you Tim.
Nice comparison, I personally feel that PIKA LABS is much lower quality in the final product than GEN 2 both in terms of resolution and in terms of the movement it provides. It was interesting to see what prompt you wrote in PIKA LABS to understand what the comparison to GEN 2 is about
Anyway, thanks for the interesting video
100% I think the "best" part was getting those kinda static Pika and Gen-2 outputs merged together. To me, that's a look at what AI video will look like in a few months...or maybe weeks. Or maybe tomorrow? It is moving SO FAST!
@@pnwviews3879 Someone mentioned that movement is better if you don't upscale. I'll need to try that out. But yeah, Pika swings wild with the movement! Sometimes a bit too much, haha.
I mean it's very cool that they upgraded it to make it look more natural. But honestly I really loved that old model. After seeing all these weird gen-2 commercials I fell in love with that type of model/style. And now it's gone😭 Do you think they might add this model back in some type of a function?
Were you in the Beta? I think that model is still working on Discord!
I’ll check it out later today- but yeah, I agree…there was something just kind of surreal and charming about it. It’s kind of like how I have this soft spot for Midjourney v4. There’s just something with a little extra sauce in that version!
@@TheoreticallyMedia Well I have a paid plan for runway. But what do you mean with discord? I thought runway is in the n
browser
Looking forward to watching The Matrix with Will Smith produced by AI!
Ha! Up there with the Eric Stoltz Back to the Future!
I can’t wait until this technology gets good enough to generate fight scenes.
I can't wait just to generate a normal scene. 😂
@@bluefernloveI can wait forever😊
I want a week long Superman Doomsday fight.
@@Didyourmomtwice An hour of motionless meditation for me, please.
It looks like it lets you add a prompt to the image from the text Gen 2 has in their prompt box? Is that After Effects you used to merge the PIKA and Gen2 outputs?
Nope, just good old fashioned Premiere! You can do it in Davinci as well!
This might be a great moment to have Hearst in Citizen Kane get overtaken by Gremlins while he sits next to the fire. I notice I'm not taking myself too serious with this tech just yet but I am focusing on how egregious I can really be. 🤣 This will get crazy good really fast - Didn't take long to make these adjustments - I foresee short films raiding AFM, Cannes, Sundance, etc. very soon.
Ha! That would be an awesome Kane outtake! Yeah, I get you-- although apparently an AI film did actually win the Jury Prize for short film at Cannes, so it has already started. (th-cam.com/video/pK7AGfBtw1Q/w-d-xo.html) But I do think the trick is just to play right now. That's going to put all of us in the position of being ready to roll in 6 months when everything REALLY blows up. It's a freight train at this point!
It would be better that these programs would be able to generate objects from scenes separately. For example, a man fishing, then the lake, then the sunset, then after everything is generated separately the software could merge them together. That way everything would come up way better.
That’s happening for sure. Fable (the South Park AI guys) are generating characters abs backgrounds separately. I mean, it’s South Park animation, so it is fairly basic, but it’s a start.
So photobashing
I wish these video generators used physics like from a game engine like Unreal; the video shifts way too much into other images to another image in the same clip.
I have to presume someone is working on that-- I get EXACTLY what you mean. The fact that you can already prompt a 3d character, and prompt a 3d location, leads me to believe it'll all marry at some point. It's kind of why I'm getting boned up on Blender right now. I think there is going to be a major breakthrough on that front in the next year.
just imagine this tech in 2yrs
That’s exactly what he said in the video
🎯 Key Takeaways for quick navigation:
00:00 🎬 Gen 2 AI received a game-changing update that significantly improved its video generation capabilities.
00:27 🎭 The video showcases examples of impressive Gen 2 outputs, with cinematic and emotive moments.
01:56 🎨 Gen 2 now outputs at a higher resolution and frame rate compared to Pika, making it a compelling choice for video generation.
02:25 🎛️ Gen 2's current process involves generating an image without any prompting, limiting direct control over the generated content.
03:49 🖼️ Gen 2 can now handle any aspect ratio for the images, allowing more flexibility in video creation.
05:39 💬 Gen 2 works exceptionally well with animated type images, producing impressive results.
06:07 🏃 Gen 2 occasionally struggles with more complex images, resulting in inconsistencies and odd motions.
07:42 🔄 Using Roop model can enhance Gen 2 outputs, improving facial characteristics and overall video quality.
08:37 📺 AI-generated personalized entertainment is rapidly advancing, with the potential for fully AI-generated movies within a couple of years.
Made with HARPA AI
Handy feature for quick chapters! Well done!
@@TheoreticallyMedia Thank you so much! Really hope this helps both navigating and discovering the video.
Off topic: I want to use Topaz AI, but it cost as much as luxury/sports car. Are there discounts on the software someone who’s near broke and a amateur like me?
Looking into it!!
@@TheoreticallyMedia thanks!
Maybe for the fixing the face we can use deepfake (although it could be somehow complicated I think) but the result might be good
Glad gen 2 upped their game. Right now pika labs are being like high school only allowing certain people to play with their tools.
Yeah, I hear you-- but I do understand that they need to do the slow roll out to stress test. I'm pretty sure they're a small team, and if you saw what happened with FullJourney when word got out about them-- I mean, it is understandable.
Although, I think Fulljourney has stabilized a bit. Have you tried them out?
@@TheoreticallyMedia I do totally understand the stress test approach, maybe it's just sour grapes on my part because, I've tried 3 times to join the beta but haven't been approved, yet still others who discovered pika well after me have been... Ah well.. Yeah I've been on the FJ discord for just over 2 weeks now. It's definitely stabilised as you mentioned, and the quality has greatly improved, but I was looking for the best image to video generator to bring my MJ images to life for a film project we've been working on.
On Gen-2 if you click upscale it has much less motion
That's a good tip! I'll experiment with that! It does makes sense, as less motion requires less resources.
*I using Gen2 since beta version*
Same! It’s come a LONG way! And that wasn’t too long ago either!
love the see example of not involving humans, like trains, bridges, jungle, etc
all doable, considering Midjourney or whatever other AI Image generator is capable of generating it. And if not, just use a photo!
It's Sean Connery and Craig, not Moore.
Oh, I can see that! I think Moore gets stuck in my head from time to time, as he was the Bond I grew up with. I mean, in the pantheon of Bonds, he was the goofy one-- It is probably nostalgia, but I always enjoyed that era of "comic book" Bond. Funny enough, the first Connery Bond movie I saw as a kid was Never Say Never, the non-canon Bond.
@@TheoreticallyMedia Never Say Never is Thunderball without the Bond theme song.
The hands tho🤣
From the videos of all creators here, it looks like these tools are mostly for generating special characters, animals, and movie like scenes or animation type scenes, all these are great but more appealing to artists. As marketers, we plan to use these tools for our marketing stuff such as actual human beings using our products or create an video advertisement that will save us from actually using a camera and actors and recording, editing and that sort of stuff. I don't think this is the focus of any of these tools right now
Check out the newest video I posted, like 5 minutes ago-- there's a section in there pointed right at you: th-cam.com/video/saZtmlJYagE/w-d-xo.html
Still too much skinwalker effects, but it's still a interesting concept taking a still image and telling a ai to move it.
Skinwalker is a great way to put it. I think once we get past the "bobblehead" (which I don't think will take long) and have the ability to direct the action via a prompt, we'll be "there"--
For the record, I do think we're "there" already, just a limited "there"...Wherever we are, there will always be a next "there"--- So I figure the best course of action is to start referring to now as "here"
"We're Here" that's my new motto!
I think we should not let AI generated videos to become boring stock video generators. That's just replacing people without society gaining anything. For me things like the "AI generated fair of the ages" showed that it should be a technology that enables people with great fantasy and a good taste to bridge the gap of having zero budget to having millions of dollars. Anyways I prefer to explore the surreal, the dreamy and not the world of stock videos.
I have no idea why you think it will only be a stock video generator. Within the next 5 years, this technology will have the potential to turn anyone with a computer into a big film studio. This is the wet dream of anyone who's ever been limited by budget, time and any other disadvantage you can think of.
Couldn’t agree more. I do think that Gen-2’s image to video is a step in the right direction. You could generate something super imaginative, or hell, even draw/paint something on your own and turn it into video.
Once we can “direct” via text prompt? I think you’ll see a lot of super interesting stuff.
(That said, the Stock Video market is about to vanish.)
Lots of time is being wasterd with this low-grade AI. They did invent VIDEO CAMERAS, you know? Also, the guy talks at 1000 word per minute. Boooooring.
I mean, its new technology. Imagine standing next to the Lumiere Brothers in 1895 and they hand you a camera and say: "Try this out"
That's where we are...only on a global scale.
Also, I drink a lot of coffee. To be honest, I'm slowing myself down in these videos.
OK, but why?
Why what?
@@TheoreticallyMedia Why make AI videos? What's the point of all this?
Really interesting
It is a ton of fun! Pika is open now too, go sign up and make some crazy videos!!
Holy shyte!!! Is this AI crap really real? Is this what Hollywood is talking about?
Oh, yeah: it gets crazier too. Check out this video, where I show the best AI film I’ve seen yet: Big AI Video News: Kaiber's New Feature & The Best AI Film I've Seen Yet!
th-cam.com/video/CmCQ2ZvB2Nk/w-d-xo.html
And that’s just a week later.
I love the tech, but other then play around with it I see no real usage for now... We need to wait for that couple of months I think!
Oh, I hear you. And it IS play right now. But I kind of liken it to playing with toys as a kid. You play with rockets and maybe grow up to work at NASA. Or you play with a toy drum set and grow up to be unemployed (zing! Drummer joke!)
But I don’t think we’re far off (months?!) from things getting super real. 2 years if you believe Joe Russo. So, keep playing and you’ll have a leg up once it all hits the fan and everyone is racing to catch up to you!
😐😮”Say”😐😮”my”😐😮”name”
I was debating closing out with “I thank you for watching, I am the one who knocks!”
You in the Indy Jones sequence makes you look like You, Harrison Ford, and Patrick Stewart mix.
I’ll take Han Picard!
It's kind of funny to me how perception of these generative AI tools has changed over time. When it was just deep dream it was all "hey look at this crazy psychedelic robot," but then as soon as robots could start to sometimes put together something coherent maybe 1 out of every 100 times, it turned into this kind of collective proud parent stage. Like because we saw how terrible they were before, the fact that the technology was capable of doing ANYTHING that looked even moderately okay was something to be celebrated. But then as they get more training the hits start to get statistically more frequent, and then all the sudden it turns into "uh oh the AI apocalypse is here literally right now hide your kids the robots are coming to kill you!" And so you've got people freaking out because they think these still completely incompetent machines are going to make art as a concept obsolete. Meanwhile you've got the people like who make the AI Seinfeld and that guy you mentioned who thinks AI movies are 2 years away and it's like...
Everyone's gone crazy. People just can't look at this stuff objectively. Everyone is just seeing what they want to. Be that generative tools that actually work, an existential threat, a way to steal art (hey, they aren't wrong, some people are just evil and will use these things for spiteful reasons), something scary, something funny, something useful, something that's going to change the world in 2 years, or 1 year, or 1 week... We're just projecting. Everyone's projecting their own personal desires and fears onto these robots that... let's face it, are shit. These are stupid, non-functioning, awful, shitty pieces of trash. Hey don't get me wrong, I'm impressed too, I think it's cool when I see one get something right, and I feel like a proud parent same as anyone else. But let's not pretend here, you can't objectively look at any of the generated video here and sincerely tell yourself that it passes the bar, you're only lying to yourself by saying that. We aren't judging AI output on the same level we judge things made by people, we're not judging it as art, because it's not art, because AI isn't actually aware of itself or what it's doing yet, it's still just an algorithm. And so people just get wrapped up in this insulated bubble about how smart and precious our child is.
We're not getting AI movies in 2 years. Robots are not threatening artists. Our baby is not smart or precious, it's very, very dumb, and extremely bad at what it's supposed to be doing the overwhelming majority of the time. To really get some perspective on this, remember that deep dream went public in 2015, that was nearly 10 years ago. And how far have we come since then? NONE of them can draw HANDS. It's not all psychosomatic, there has been a lot of progress made, but like... the technological leap from this wacky garbage to genuinely useful generative tools is a hell of a lot further away than "just around the corner." It's going to be at least 5 years and probably won't be for another decade, really. And that's just for TOOLS. Remember, none of this is even real AI yet, the actual goal of making an awareness we can communicate with that can understand us? The way things are going, we'll probably go extinct before we even get close.
Some great points in here, and I actually agree with a lot of them! It's funny you mention DeepDream, I haven't thought about those wild psychedelic "dogs sprouting eyeballs" videos in ages...I remember thinking they looked like one of those old 3D stereogram images on mushrooms. Hmmm, now you've got me thinking I should try to generate one and animate it.
Anyhow, back to your point: Ultimately, I think generating AI Video (at this stage) is a little like hip hop sampling. You're creating raw data (generated video clips) and then you have to sort/remix/chop/retime everything into a form that achieves your desired result. I always stress that there needs to be a human element to direct AI Generation-- you called it: They're tools. We create, they generate a mess of non-matching puzzle pieces, we assemble it into something new.
There have been a number of "Hollywood is DEAD!" proclamations from Chad sector. I'm not one of them (I think I'm too old to be a Chad, anyhow?) I don't want AI Video to compete with Hollywood, I want to let it be its own thing, its own medium. I don't want AI to be able to ingest a film script and output a 2 hour movie...because, well-- there would be even more crap movies.
I love this era we're in-- where artists are playing, and creating weird and wonderful stuff just because they can. Check out the latest video I did, which showcases a number of them when you get a chance: th-cam.com/video/saZtmlJYagE/w-d-xo.html
When you have a chance to watch it, I'd would love to continue this conversation over in that comment section, if you'd like!
I get why the actors would want to strike but at the end of the day, life is change. There will always be a place for actors imo but maybe it will have to be on the stage or something else. If people are not willing to change, it leads to stagnation. We all know what happens to things that stagnate.
I primarily used Gen-2 for this, and a ton of editing. th-cam.com/video/DDx9lpPUBYc/w-d-xo.html
nice work! I think where we currently are: Horror and Surrealism are perfect genres. Which is great for me, because I love David Lynch movies. Nice choice on the Black and White as well!
Yeah I managed to get just the reference image working but it's weird that it doesn't let you inform the action with a prompt. They really ought to focus on the animation part, anyone seriously considering using Gen 2 won't be using it for its image generation yet
I presume that prompting is coming down the line. That seems to be the AI way: Launch and then we'll figure it out/improve as we go along. Haha...It is messy, but fun.
I could swear I commented about text-to-video generation on your channel a few days ago, and bam! The pace is mind-boggling. The possibilities are endless. 1:01 This, is ridiculously good! We've gone from dog surfing to a few seconds of light emotion-tugging videos in what, a month? The analogue aesthetic actually works incredibly well with the scene. Your videos are always entertaining and informative, keeping us updated. 32 _thousand_ people are turning to you! So no, not _just_ a YT'er!
But yeah, Russo is right ~ 2 years. 😆
Yeah, I think you did on the Pika vs Gen2 original video. That was...oh man, I can't even begin to express the sigh when I posted that video and within hours Uncanny Harry was like: "You gotta update your video, Gen2 just dropped a major update"....
Haha, thought I was sleeping that night, guess not!
Awesome video, as always. Thanks Tim.
Really glad you enjoyed it!!
This video made me wanna fish and drink. Gen-2 is lookin' good, man...
Make sure to hydrate! That sun will get you!
@@TheoreticallyMedia Or - Fade Out!
awesome video as always! thank you!
Thank you so much!! This one was a lot of fun to put together! Really happy you enjoyed it!
WOW!
Right?!! Super crazy!