Actual AI Text-To-Video is Finally Here!
Вставка
- Опубліковано 18 вер 2024
- We actually have a working text-to-video model where you enter a text prompt and it will attempt to generate a video from that text. Here's a breakdown of how you can use it right now.
Hugging Face Text-To-Video: huggingface.co...
🛠️ Explore hundreds of AI Tools: futuretools.io/
📰 Weekly Newsletter: www.futuretool...
😊 Discord Community: futuretools.io...
🐤 Follow me on Twitter: / mreflow
🐺 My personal blog: mattwolfe.com/
🌯 Buy me a burrito: ko-fi.com/matt...
🍭 My Backgrounds: www.futuretool...
Outro music generated by Mubert mubert.com/render
#AI #NoCode #Futurism
P.S. I have been able to generate videos for free. As soon as this video went live, I stopped getting the error messages on Hugging Face and was able to generate videos for free. So, if you're patient, you can do this without paying a cent... It might just be a little bit slow. Video generation was averaging about 5-minutes on the free GPU, vs the 1-minute when I duplicated and paid for an upgraded GPU.
Like I said 1-3 years before Ai Generated Movies is a reality
Just trying the free one now and I still get regular errors, but I was just able to get one to start processing. Estimated time is giving me 10 minutes. Awesome stuff, thanks for covering it!
@@domehouse79 based on the rate of development progression for Stable Diffusion, it's very reasonable to assume within the next 6 months that we will see full text to movie generation, or at least concept short films.
@@domehouse79 well that is optimistic but I can see it. i really thought something like gpt-4 would be 60 years away a few years ago
Maybe I am not finding the link you mentioned because I am trying to read it on the phone. Is there any way you could repost it?
Thanks for all you do.
Ken
The speed at which this AI development is moving is simply ... mind boggling.
reminds me of early 2000s with explosion of peer to peer technology and filesharing, it feels a bit wild west, copyright and other ethical questions take a backseat to radical innovation by amateurs and experts both contributing... it was a huge era of progress, but there will be a blowback and then some stabilisation.
It’s factorial because AI is iterative and each generation feeds into the next one by helping train it, helping produce the training data, even helping design the AI specific next gen hardware
@@StoutProper this could be a bad thing as well, no? Because then we are essentially making a copy of a copy this way so I think if it worked this way we would be getting less accurate over time but thankfully that’s not what is happening
Ive been saying this. AI progress doesnt follow Moores law, it follows Dennis' law.
Not a doubling every 2 years but power of itself, its exponential.
Lets say u start ar 2 of "power" in two years u have 4. Then 16. Then 256. Then 4 billion something then like 8 quintillion
its more about computing power than it is about AI because most of this is not really an AI in sense most of the people think it is, they are just huuuge trained models and databases that are handled with the extreme speed and precision and those operations cost millions and millions
The pace is moving SO FAST, it's actually unreal...but we're keeping up together thanks to all your hard work! Thanks Matt!
It’s factorial
It took 4 minutes to make a 2 second static image unrelated to the prompt.
I really like the thorough step-by-step style in which you showcase those tools. Especially when you don't cut out an error message but explain why it happened and how to deal with it. This really makes those tools but also the infrastructure to run them much more approachable and easy to dive into. Keep it up!
"These are completely original creations, these AI models don't copy anything...
...except for the Shutterstock logo"
Way too early for me to play with this one for now. But I like what is going to be possible. For me at present, I'm playing with ChatGPT V4, and actually using it for hints and ideas and bits of source code and other things for work too. Clearly 2023 is the year AI can't be ignored any more, and is going mainstream in so many places.
Yep to humanity's detriment.
Me when images that will always look blurry when you zoom in will be the detriment of humanity
Tell me when it starts generating hands and ears properly and we'll talk
@@Magicwillnz Hundreds of millions of billions of people are going to have their lives careers and futures utterly destroyed. If there was a 'bleakness' measure for the future it went into the stratosphere. So much human suffering is coming.
How has it been working out for you?!
@@derpherp7432 The only reason these people will have their lives destroyed will be because of corporate greed, if we can automise all their jobs, then we as a human race have come far enough to give people a livable income without having them working. So you should be worried about how greedy humanity's top percentage will be
Move over Steven Spielberg, we have AI-generated movies now! 😂
Steven Shutterstock and George Lucasimages are definitely the up and coming directors now, baby! 💪😎✌️
James CamerA.I.on will be crafting Avatard 3 through 12 in 2026. Award-winning graphics.
🐲✨🐲✨🐲✨
Spielberg is producing his own AI platform that will make CGI more accessible, its called wonder studios
@@Novastar.SaberCombat LMAO
@@Novastar.SaberCombat avatard 😂
D'oh!
Once the technology matures, this will definitely open some interesting doors in the world of adult entertainment.
The internet is for Porn!.... Oh I'm sorry I misunderstood what you meant, I thought you talked about musicals XD
oh yes. rule34 will take various weird categories to a whole new level
Already happening with anime rule 34
Child 🤡🤡🤡🤡 would be real 1st
Mmmmhmmmm
The temporal consistency really impresses me. This is quite good.
it is not good at all, I sat with it yesterday and tried to generate nude women and sure sometimes it work but most of the time they morph in weird ways that make no sense at all and their faces are always completely messed up too. In a couple of months it might be worth using though
@Tinselfect ykno that porn sites get more visitors each month than netflix, amazon and twitter combined. It's NORMAL
What part? I agreed with Matt, it's SOMETHING just not very impressive at the moment, give the DEVs some time. Sure the teddy bear was cool, but everything else --sucked-- left me feeling non-plussed.
@@Danuxsy it says that it's not allowed to generate pornographic content
@@ratdoctor well u can gen nude women that's for sure lol but it isn't good at all so yeah
Cool stuff. For the time being, I'd stick with one of the options like Kaiber. I do hope this improves though, and I'm sure it will.
Fr AI is growing so fast, imagine what it could do in a few months 👀
lol. How long do you think it will be before shutterstock files a lawsuit......
I have no idea if Shutterstock gave permission or not. I imagine they'll be retraining on non-Shutterstock images soon but who knows.
The scary thing is that it's completely ingrained into the videos all around, actually coherent text.
Literally just the beggining of the week and we already have this for this week in ai
I remember back in the early 1980's when I showed someone what I could do with a Vic-20 in Basic, they were really unimpressed. I went on to convince the community hospital I worked at to begin moving towards the use of computers in the latter 80's. That early individual did eventually come back and say they were sorry they did not "see" what I saw was going to revolutionize the role computers were going to take over the next few decades. It's your enthusiasm, Matt, that keeps me coming back. Having Vision for AI is being able to see what the early stuff will eventually grow into as systems get fine-tuned and improved. That's true for EVERYTHING, especially when dealing with computers. Viewing these early attempts do not leave me unimpressed. I feel like a little kid coming in to the living room on Christmas Morning and seeing all the wrapped packages that weren't there when I went to bed the night before.
All these Ai generated imaginary wasn't even a topic 2 years ago, let that sink in. In 5 years it will be so advanced enough it will be capable of detecting every single pixel for possible artifacts and aim for ultimate photo realism. Nobody will be able to tell its fake.
Now that we have the first open source model for txt2video, I'm excited to see how people will manage to fine-tune and create merges and loras.
Exciting stuff ahead now that smart people can iterate off of this!
Thanks for the constant update in the ever-evolving AI world, Matt. I'm still mortified at the speed at which it's improving, makes me wonder what fun or horrors await us. The constant thought on my mind is what the inevitable societal response to these programs will be and if it will end up exclusively in the hands of corporations. I can see AI's democratization being led by something like Alpaca or LLaMA say ChatGPT's openAPI had the plug pulled, but I think we just have to wait and see what direction the world takes us.
You'd imagine a truly advanced A.I would dominate the corporations rather than the other way round.. it feels like the equivalent of trying to hold a tiger by the tail and profit from its nature..
that's not to say I disagree with your prediction, merely any corporation trying to shackle such a beast to their advantage may not get the docile creature they were hoping for..
for now these systems are disparate and weak little things but as time passes and we see them grow and combine, once that growth spurt has passed I suspect the time for misuse by corporations and governments may have already passed them by..
Yea, I have a theory that I think that this fear is what is sounding the panic alarm button, and whenever this happens, then corporations and governments try to act as a controller, and they become middlemen, and then we have to pay to access these techs..
All this AI stuff is amazing, but we can’t believe anything we see, hear, or watch as being real from this time forward.
People are going to dial-in to this thing like the matrix. I hope some of us stay in this reality.
@Madalin for now
You must be blind to not see if it’s real or artificial, come on
Appreciate the hard work and keeping up with it and sharing it. 👍Yes, things are moving really fast. Yes, this was underwhelming. Yes, we'll be making some cool videos in less than a year.
No, we will not get enough sleep!
_"we'll be making some cool videos in less than a year"_ don't you mean we'll be pimping our GPU's to make money without any personal expression, artistic interaction, what the crowd calls PASSIVE INCOME?
You THINK? You THINK that's going to happen? What you call THINKING isn't what happens when I THINK.
Sorry, that went dark, I was just expressing myself authentically in a moment. It's not personal and surely this IDEA is completely unrelated to Token Currencies... not like the real ones are more than dreams/shared delusions.
The speed is crazy . Things are moving fast , it's insane
The fact that any of them look good is an indication that it won’t be long. The phrase Matt used “trial and error” just means it needs a bit of training. Very exciting stuff! What, maybe 6 months and it will be great.. maybe?
just spam itll work after a couple seconds of trying and take about 5 min no need to buy anything in my experience if you got a little time on your hands. have fun^^
We can convert our dreams to text to video.. Finally 😊
Thanks Matt for giving us a heads-up on this new technology. I was able to get a few free videos created. Probably the best one was "Flamingos on a beach running around the beach". Shutterstock was in the middle of this one, but not on the others. Can't wait to see how this develops...
I love Text-to-Image but this video thing still looks too experimental for my purposes. But I am also sure it won't take long for the video models to improve and they will surely be helpful for me in the future. Nice informative video.🤩
I got a 2 second static image that was unrelated to the prompt after waiting for 4 minutes.
This is basically what text to image looked like when it first was developed. Give it a couple months and it will quickly improve.
AI has to calm down. I can only practice,program and keep up with so many artificial things at once💔
Same, it's all so fast and info how to use it all seems limited or very technical too. I think we're gonna have to buckle up 😯
@@kewlnes987 I started mastering a couple then hear 10 prototypes and a whole new version then cried lol
The tools are getting more notoriety than what people can do with them outside of corridor crew, the colorado state fair piece, and a Linkin Park video
All this Ai generated imaginary wasn't even a topic 2 years ago, let that sink in. In 5 years it will be so advanced enough it will be capable of detecting every single pixel for possible artifacts and aim for ultimate photo realism. Nobody will be able to tell its fake.
Thanks for keeping us in the loop during this crazy time
I just realized that UA-cam is the Netflix of the future. like, high quality, long-form, live action tv series and movies people will make in their free time. it's going to be an insane paradigm shift. Imagine every fanfic you've ever read, being made into real content as well. there will be lawsuits once the fan videos start repeatedly being better than the official shows and movies.
@dave smith oh no doubt. paramount already legally basically made it clear that you can make star trek fan films as long as they are not GOOD. the quality of fan films is gonna go up a LOT and actually start chewing into the profits of these ip owners. so there will be a legal firestorm that will happen.
8:54 is actually kinda of terrifying hahaha
It is important to note that, while the Shutterstock watermark on the output obviously does indicate that the model was trained extensively on content that had Shutterstock watermarks in it, that really does not directly imply that the training infringed on Shutterstock's rights in any way, much less that the output represents some kind of straightforward copying process by which it just re-outputs training data that was input into it. There is a lot of confusion in this in the arts community right now. A perception that generators are just simplistically retaining and splicing-up existing content in highly mechanistic ways, without anything interesting going on from a technological-innovation perspective.
I want my goddamn Firefly Season 2 made by AI
Wasn't expecting this comment
*Y E S*
Just because it’s new and there’s a novelty factor, doesn’t make it less atrocious. It’s utter garbage. And you PAID FOR THAT!😂😂😂😂😂
I’ll come back in 5 years when it’s made a bit of progress.
That would be a lot more interesting to train those models on 3D generated videos from gaming, generating hours and hours without licence
Yeah I wonder why they don’t use games to generate the video
It's coming soon, I'm sure of it!
Someone's training that right now, I'm sure of it
but why would they use gaming content? they hardly look realistic
The hours of video are not useful without detailed descriptions of the videos. Also, the imagery in a game is often very limited to a particular scope and art style, so the ai would only be able to generate that style.
Jesus christ, already? At this rate, in less than a year we're gonna have AI chatbots that can generate video clips dynamically as the action progresses to display what's going on
The day when you can take a screenplay and manufacture your shots with it will open up a new universe for creators and Hollywood could become virtually obsolete.
Every AI video with eating is the making of a horror movie lol
When does everyone reckon the first ever feature-length movie entirely produced by AI will be released? My bet’s on 2027!
For the record, I mean entirely produced by AI. The script, the video, the soundtrack, everything.
@Divergent Integral When you put it like that, it makes you realise just how unbelievably quickly technological progress has accelerated in the last century. And to think it’s probably only going to get quicker and quicker as we approach the singularity…
Next week! lol
@@AG_before I was just about to say that haha
Movie in a week, metaverse in a month, robot takeover apocalypse next Fevruary
Hah. Human arrogance. Try the twelfth of never. Ever.
Bro, things are getting really cool. This could put Major Movie studios out of business unless they become UA-camrs.
I mean the model is only 1.7B parameter's, pretty impressive for its size, and considering its dataset seems to be limited to Shutterstock.
This thing was expected to happen with the pace AI updates were coming from few days. Exciting!
People are trying to give the AI the most crazy possible ideas to check results. I believe in the coming years a boom of surrealistic art is going to take place again, but this time will be digital and generated by computers. If this will be considered art? Well, in the same way people use the word art today, it might be considered art for some. I do still like to separate art from commercial products designed to sell. Anyway, I still believe this is going to be a huge relieve for commercial work, living the time for real art, to get free from those superficial aesthetic decorative purposes, to those who want to get deeper in the real meaning of making art. And AI is already used from real artists to make their works
Can't wait for a text to 3d ai generator for producing 3d models of characters and environment landscapes
i still remember people saying a little over a year ago "imagine if they make Ai generated videos in 10 years" i wouldn't be surprised if we see indistinguishable Ai videos in 3 years. All these Ai generators wasn't even a topic 2 years ago, it came out of nowhere
I feel like in 3 years something like that will seem like nothing in comparison to the stuff we'll be able to do
I remember playing with a text to image and put in my favorite artist, Gustave Dore. His name, winter, death... oh boy it made some chilling stuff.
Tried it, it's a bit crap right now. You're right about the cherry picking. I'm sure it will get better
It’s only 1.7B parameters and it’s pretty new. It will get better for sure
8:05 Gosh if those are the best, I think there’s room for improvement. 😂
9:03 Yours is better because at least the bear is running and not magically floating along on his bottom.
This A.i. evolvement is going so fast its unbelievable
2:36 GRRM would get a real kick out of that lightsaber
Invideo is coming out with an actual text to video where you just input an idea and the video is created and everything is copyright free. You can also add in your own things if you want but you don't have to.
text to reality technologies will rapidly advanced over the coming years, eventually we will have text to 3d object (create any 3d structure, a castle, a city, etc..), text to 3d world (create entire 3d worlds, like a whole aquatic world), text to movie (create whole movies), even text to 3d object generator (combined with a universal constructor, you can literally create any item you want, say a diamond with wings, etc..), and this is probably just the beginning, i'd say endless text to reality technologies will be incredibly advanced in the 2030s
I always thought Shutterstock watermarks were pointless since people just use the image anyway and ignore the watermark, but it turns out they were a decade ahead of the game.
better yet, use Photoshop to remove the watermark, or Ai to remove it 🤣
Wow that’s amazing!
Wow, the extremely fast progress of AI must be because the exponential (accelerating) development of technology that Ray Kurzweil has talked about.
It's like the tech explosion with pcs back in late 80s-90s, we are hitting that sweet spot right now with A.I.
@@Elburion I don't think we've ever started to get to the sweet spot. These tools are still being developed primarily by humans. When Ai reaches the point that it surpasses human's in programming capability and complexity, that is when this technology will accelerate exponentially.
Ai will design better Ai, and better methods for making Ai, and do so at a rate humans could never manage. Every other field of technology will get dragged along in it's wake as it rockets past anything anyone imagined just a year ago.
@@Goodgu3963 exactly and agree with you, ai will design faster better computer technology, which will then be used to further design more advanced ai, we have opened Pandora's box and now that so much is open source, I don't think there is anyway to close it.
The future is here, brother! Every day is a mind blowing day in Ai...
it looks like those video samples that came on cd's to use in arts. In the past.
Very promising. The AI is evolving extremely fast.
Voice to video will be the true revolution
I find all of this very cool. Thank you for your hard work.
I thought a good midjourney prompt for this would be: "a hyper-realistic photo of a process server dropping off the shutterstock lawsuit paperwork to a hugging face zombie at the front desk" with some weight on zombie, maybe?
matt, i just LOVE your on point criticism. no sugar coating, you say it as it is. respect for that.
I can't wait until we get the anime version of this so we could feed it prompts from mangas and complete the unfinished animes
Just imagine the possibilities of using ai in physics in the future. If we ever come close to a theory of everything, i doubt a human will be the one that finds it.
Thanks for your hard work! I just discovered your channel yesterday, and I'm hooked. I'm a designer and I'm always looking for useful tools for design such as creating wireframes or layouts based on content that I have, or conversely, creating content based on the layout that I have. This could be useful for graphic design, web/app design, etc. But I would also love to see other industries such as architecture, landscaping, product design, or even design systems or production systems. Thanks again
Yea… I’m waiting for text to movie! Imagine typing your dreams and having the AI process it into a legit movie
2:56 "hugging space" im falling down into the grave rn 💀
In five years tops we'll be able to generate a manuscript which will be voice synthesized and mapped to a full feature movie, all in the same tool.
This AI is absolutely phenomenal and such a great thing for humanity. I love technology.
I feel like we'll get actual cool text to video AIs the moment Microsoft and OpenAI will get interested in it. As much as stuff like RunwayML gen-2 is cool I feel we'll get really impressed the moment big companies will release their versions of it.
X years from now (at the pace things are going I have no clue when), Hollywood will be obsolete. You just tell your tv what kind of movie you want to watch and it'll create a masterpiece just for you.
Therell defintely be means of curation from on high trying to control what people generate.
"cute teacher with massieve jugs getting swarmed like that guy in Jurassic park by 18 year old students"
Thanks Matt for turning me on to trying SD w Deforum. I have been at it for 10 days now and can now create 3D videos consistently.
Here’s to the buzz not fading out. Let’s all just keep needing out into infinity and beyond. At this pace it will not even take half a year before this is high quality.
finally I have been waiting for this.
And the Best Picture Oscar for 2033 is... again a computer.
I’m watching this video in November 2023 and it looks like this was done in March 2023. I’m guessing the technology has accelerated so fast that if I check your channel now you probably have something far more impressive for the results of this tech. It’s moving that fast isn’t it?
Of you’ve used stable diffusion you’ll know that seed changing can be a powerful tool as well.
Must try this. I've been creating images on Stable Diffusion ever since Matt's hugging face demo a month ago and I have found that sticking to the same phrase over and over seems to create a wide variety of images and quality. About 1% or 2% of images are really good and 5% to 10% are ok, though yes the Shutterstock watermark is sometimes a problem but not nearly as bad as on text-to-video
This is what dreams look like if they were interpreted into digital form
Man your content game has been ON POINT ever since I found your channel. Your videos have helped me a LOT!
This is only just the beginning. When the dust settled we will see who is the true King of AI will be.
The way these text to video models are trained is interesting. Personally I would opt for GAN networks over stable diffusion
Using a TiVGAN network ( text to video) would generate far better quality. Text to video Generative Adversarial Networks with GPT tockenization mapping models can make this stuff explode. I'm talking about generating a video frame by frame then scene by scene. You could potentially create an entire movie this way.
Rest assured, this stuff is coming. The only thing holding us back is computing power. These models cost millions to run.
Your running teddy bear was actually better than theirs, might wanna shoot that over to 'em for their ad! lol
If this gets perfectionized, we are able to recreate our dreams
Just yesterday I was wondering how long it'd take for programmers to create a text-to-video AI. Can't believe this is happening so fast. Where will we be in the next 5 years???
@@kotcraftchannelukraine6118 bruh calm down. U r giving me goosebumps🙄🙄
its been out for a year now bro
@@kotcraftchannelukraine6118 i know but its not worth a try in its current state nt if it becomes d way u described which i believe is not very far in future it would instantly replace the entire entertainment industry.
does anyone knows how to download and run this locally
I tried it with a prompt saying "people walking in the street" but it took like over 300 seconds, and it generated a 2 second video, but with a catch. The catch is that it had something that says "shutterstock" (a stock image website) even though it was AI-generated.
Did you watch this video?
There is also Genmo and Kaiber who does text to video with similar results. And hopefully today Runway.
I look forward to watching and learning from what your videos are about. Technology was something I would never say I was good at. And as I am still not great, what I learn here has truly helped me look like a ROCKSTAR!
Geez, can't wait for image to image, we can all be movie stars. For better or for worse. But I wouldn't complain about a Big Lebowski 2.
Those puppies look like their family tree is a ladder 🤭
thank you for the video , i learn a lot from your channel , i really appreciate , about the watermark to avoid it because there no negative prompt , you can use " --watermark , --logos ......" in positive prompt , to avoid anything use two --
I look forward to the day when there is a text to video that is worth my time. Thanks for sharing it saves me a lot of time.
I think instead of trying to train a different model it could be easier to train an advanced image generetor like midjourney for example.Although this solution would make it increadibly slow but at least it would have more image quality. On the other hand main thing video ai's have problem is constancy maybe teaching them how 3d space works and teaching them what is volume would be more helpfull than just blindly showing them different videos,images ad videos
Imagine how much this technology will advance in, not even a couple years, but just six months. We saw it with chat GPT, and mid journey.
well this is interesting, i think this tech will take some little more time then text to image
Today us a very frustrating day. Open AI GPT4 isnt responding to ANYTHING. Now I find a text to video finally exist & that does work either. Very frustrating day.
You might as well use stock videos instead of this. But hopefully this tech will develop fast.
I'm keen to see where this goes :) exciting! 😮
In 10 years, UA-camrs will be making avengers level movies using text to video. This shir is getting scary
hopefully they will have more interesting story lines than current DC derivative junk
@@markwalker8374 I imagine so. Because, we will be getting stories written by people from all over the world. Not just stories written by those few big media teams
shutterstock must be so pleased LOL
Had to laugh at the shutterstock watermark because all the stock photo websites steal others images and claim them as their own.
It looks pretty interesting but it's still early so we will see how this pans out.
The perpetual embossed shutterstock is the chefs kiss. Looks like you'll still have to pay someone on fiverr just to bust those pesky copyright watermarks. 🤔😏