I want to see both installs but definitely the Blender one. I wish someone with build a more user friendly app for these open source generators but then I guess you would need a team like Blender has. Thanks for all the great AI content Matt
Honestly cant believe how finding your channel today has completely expanded my mind in so many ways...been on this rabit AI hole for few months and ur channel covers things in best diff way and covers all i been looking for thank u
Wow! ...that sounds good and promising! By the way: I really like the mutating incoherence of SD Deforum: Of course not in every context, but with fast psychedelic music (PsiTrance, Progmetal etc) the Deforum style fits perfectly. Thank you for all your great and quick updates! ( I love all the Lemon testings very much!)
num_steps: is like in Stable Diffusion, controls how detailed is the image. It should be at least about 20 and bigger values generate more detailed but not always better scenes. num_frames: is responsible for video length.
@MattVidPro Matt Wolfe also pointed out it would've taken 100s of generations to get those tiny 2 second ModelScope videos to look that coherent. I've tested 100s myself and it takes a lot of work, prompt editing, etc to get it looking just OK. Good for funny TH-cam AI videos, but nothing else.
@@davidpaxton6402 Exactly! I used it to make a funny AI Conan O'Brien video, the absurdity of the results does actually make them pretty hilarious. It's definitely not for serious video stuff yet.
@@AI_ArtShow Hi it's David Paxton here again from my other account. I checked out your channel and love your AI videos. Very creative and funny! May I ask with the Trump video: are the images from MJ or Leonardo? Also which tool did you use for Trump's voice? I've looked at many AI voice generators using well-known characters, but hard to find a good one that isn't expensive to use or sounds authentic. I'm doing parody Ai videos too. Thanks. You also have a new sub.
Sadly I couldn‘t get it to run, but the watermark still would be a bummer and makes it less usable. So I will wait until it comes to automatic1111 😅 But we are slowly getting there. I am thrilled.
Stop crapping on Modalscope! ;) I LOVE some of the creative solutions to prompts I would come up with. I am totally smitten with GEN-2 as well but I can't make my actors do much but stare into space most of the time. Thanks for the head's up, I am hoping this will be the best of both worlds.
Since this is based on ModelScope, I think you should be able to find that there is basic image coherence at 10 steps per frame. Also, use 8-10 fps and interpolate with FILM or similar tool
It crashes often in Google Colab trying to allocate memory. I couldn't create videos longer than 4 sec and sometimes just 1 sec. But I need to try to change resolution, it is not in the editable params but in code section.
The same here. It's very frustrating. It only does about a 1/4 of a second on the default settings. That's not video. it's a still image lol.. So I lower fps to 15 and try 30 frames and it still runs out of memory. I'm also on Colab Pro. Which settings did you use for 4 second videos please?
Don't get me wrong, iam excited to see how this AI goes, but the only thing i want is being able to do at least 3 minutes videos in like 10 minutes not 1 hour.
This technology is very recent and the ones available at the moment are just prototypes. I wouldn't be surprised if within two years we will have photorealistic AI generated videos at real time.
I'm just wondering how long until we can plug in a clip and ask for changes. Not that I want to make my own fixed version of the starwars sequel trilogy or anything 😇
We are going to have very serious legislature to deal with soon regarding AI and I would not be surprised AT ALL if this results in less freedom and rights for us. The problem ultimately is that as AI gets better it is harder to know what is fake and what is real. Because of this we will end up hurting many people in a failed attempt to regulate and control the AI. What it will eventually boil down is deciding upon control over the AI, or individual freedoms. If we opt for freedom we will be constantly attacked by AI, meaning people using AI, and if we opt for control then we will have internet regulation similar to the great firewall of China. Maybe I am wrong, but I don't really see another option apart from AI becoming sentient :)
I wonder if this can also be used to stabilize v2v. So far I'm getting only trash with persons changing their outfit constantly and likewise unstable landscapes and backgrounds. Looking forward to try it locally either way as I'm not a big fan of mega corp cloud stuff or discord bots to say it in diplomatic way.
Does this involve me going to github and filling up my hardrive with garbage that doesn't work and just throws off a million error messages? Because I'm already pretty tired of that stuff
Nvidia sent you a RTX 4080! WOW! Awesome! I guess they consider you promoting them, since we want to do things that require an Nvidia GPU. My only reason to upgrade to a newer Nvidia GPU, is for faster AI generation, which means I'll need the best of the best, so wise move on sponsoring you!
The next will be and ai DAW vst au voice generator where you can imitate any voice or take vocal recordings to a higher quality, the maker of this will be the next level
Guys is there a way i can find the name of that *satisfying car game* used to boost screentime in short videos and tik tok videos? Any img search method you know?
I'm sure under the hood it's cool, and in a year or so it will go from Potat0, to Potat5head, at which time it will be cool. I think this current stuff is pretty unimpressive in terms of the other AI stuff going on. But hey, thanks for covering this.
I dunno, I've been getting some really coherent high res video from Gen 2 lately, been playing with the CFG scale a ton, and a clear and concise prompts with the right keywords usually gets you there. Adding a driving image helps a lot too, but lately, I've been getting sick results without the need of an input image :)
Actually the free versions not really gonna work for most, as it runs out of memory pretty quick, even on a 16gb card. This one's definitely gonna need 20 gb or 24 gb
Matt you tend to waffle on a bit - you should have segments on the red TH-cam line so we can see each part of the video and go to it - like setting up in collab.
I hate to keep bringing this up to countless youtubers, but it's honestly a pet peeve of mine. Please mute your discord when recording videos, it's not only annoying that it's in so many youtube videos now, but it's also confusing for people who have discord as well and think they're being messaged. Not to be a dick but just a humble suggestion.
@@omnigeddon Whenever I can. And when something tackles my mind I can also do some research on my own. We are sort of cooperating. Nothing too special due to limitations you have just mentioned, but trying to test as much as I can.
Make no mistake, aint nothing beating midjourney in terms of quality. All newcomers continue to fail in comparison...midjourney is lightning in a bottle...
CORRECTION: the square example vids are from the OG modelscope. Not the new model.
Hey please tell me where is the google collab link?
Second 🥈🤝 to comment here.
@@theaieyee yeah, that's pretty cool to get continuous free GPU, if it is continuous. :3
please make a 20 minute video of 14:15
I'd say as long as coherence is maintained, we don't need high fps. We have interpolators for that, once you have chosen a generation you like.
Same with resolution. 480p would be totally fine if coherence is on point.
Kudos to you for always being the first to introduce new tools!
I want to see both installs but definitely the Blender one. I wish someone with build a more user friendly app for these open source generators but then I guess you would need a team like Blender has. Thanks for all the great AI content Matt
Noted!
Honestly cant believe how finding your channel today has completely expanded my mind in so many ways...been on this rabit AI hole for few months and ur channel covers things in best diff way and covers all i been looking for thank u
Wow! ...that sounds good and promising!
By the way: I really like the mutating incoherence of SD Deforum: Of course not in every context, but with fast psychedelic music (PsiTrance, Progmetal etc) the Deforum style fits perfectly.
Thank you for all your great and quick updates! ( I love all the Lemon testings very much!)
num_steps: is like in Stable Diffusion, controls how detailed is the image. It should be at least about 20 and bigger values generate more detailed but not always better scenes. num_frames: is responsible for video length.
can't wait to have this as an extension for automatic1111, that'd be where this will blow up
Those example videos are from the original Modelscope example i think. I remember Matt Wolfe going over these exact examples like a month ago.
Oh really? Confusing that he would put them there. I should have paid more close attention
@@MattVidPro There is really no indication that they are so not really your mistake, it's on the person who posted them. Just letting you know :)
@MattVidPro Matt Wolfe also pointed out it would've taken 100s of generations to get those tiny 2 second ModelScope videos to look that coherent. I've tested 100s myself and it takes a lot of work, prompt editing, etc to get it looking just OK. Good for funny TH-cam AI videos, but nothing else.
@@davidpaxton6402 Exactly! I used it to make a funny AI Conan O'Brien video, the absurdity of the results does actually make them pretty hilarious.
It's definitely not for serious video stuff yet.
@@AI_ArtShow Hi it's David Paxton here again from my other account. I checked out your channel and love your AI videos. Very creative and funny! May I ask with the Trump video: are the images from MJ or Leonardo? Also which tool did you use for Trump's voice? I've looked at many AI voice generators using well-known characters, but hard to find a good one that isn't expensive to use or sounds authentic. I'm doing parody Ai videos too. Thanks. You also have a new sub.
How can u install locally? Google collab just keeps giving me errors, can't generate anything with it
same here
Sadly I couldn‘t get it to run, but the watermark still would be a bummer and makes it less usable. So I will wait until it comes to automatic1111 😅 But we are slowly getting there. I am thrilled.
Stop crapping on Modalscope! ;) I LOVE some of the creative solutions to prompts I would come up with. I am totally smitten with GEN-2 as well but I can't make my actors do much but stare into space most of the time. Thanks for the head's up, I am hoping this will be the best of both worlds.
Why doesn't the huggingface link work? On the huggingface page, you cannot actually generate any video. Then whats the point of that HF page?
You spoke for three hours of hot air without ever looking at the camera once
Since this is based on ModelScope, I think you should be able to find that there is basic image coherence at 10 steps per frame. Also, use 8-10 fps and interpolate with FILM or similar tool
It crashes often in Google Colab trying to allocate memory. I couldn't create videos longer than 4 sec and sometimes just 1 sec. But I need to try to change resolution, it is not in the editable params but in code section.
The same here. It's very frustrating. It only does about a 1/4 of a second on the default settings. That's not video. it's a still image lol.. So I lower fps to 15 and try 30 frames and it still runs out of memory. I'm also on Colab Pro. Which settings did you use for 4 second videos please?
Yes please do a video tutorial on how to run this on our PC. Would love to get my 3090 cracking with the 24gb of VRam.
Tech like this could eventually be used to transform any book into a movie.
*the Bible lmao*
I know there is some bible movies out there that already exist but yeah, that would be fun
@@AltKaxREAL that's the last thing we need
@@literalsarcasm1830 hmm, any good books you think that could be interesting to bring to life?
@@AltKaxREAL the Spinward Fringe series by Randolph Lalonde
Thank you for always bring us the most updated and fresh contents
the guidence_scale at 23 look like it is waaaaaay too high. (resulting in blown-out colors)
Try like 4 to 7, 16 at the most.
Don't get me wrong, iam excited to see how this AI goes, but the only thing i want is being able to do at least 3 minutes videos in like 10 minutes not 1 hour.
This technology is very recent and the ones available at the moment are just prototypes. I wouldn't be surprised if within two years we will have photorealistic AI generated videos at real time.
Matt any updates on this video since its a year old?
Had no time for AI news for just 2 weeks and now i need to watch 12 Videos. AI is running fast.
I tried it with the prompt "city traffic" but it's just random lines with different colors
I have problem downloading and viewing the mp3 file, if I download and play its just blank, how can I solve it
It looks like image generation from a year ago. I hope it will evolve, too!
would love a tutorial on how to do a local install mate
I'm just wondering how long until we can plug in a clip and ask for changes.
Not that I want to make my own fixed version of the starwars sequel trilogy or anything 😇
We are going to have very serious legislature to deal with soon regarding AI and I would not be surprised AT ALL if this results in less freedom and rights for us. The problem ultimately is that as AI gets better it is harder to know what is fake and what is real. Because of this we will end up hurting many people in a failed attempt to regulate and control the AI. What it will eventually boil down is deciding upon control over the AI, or individual freedoms. If we opt for freedom we will be constantly attacked by AI, meaning people using AI, and if we opt for control then we will have internet regulation similar to the great firewall of China. Maybe I am wrong, but I don't really see another option apart from AI becoming sentient :)
I wonder if this can also be used to stabilize v2v. So far I'm getting only trash with persons changing their outfit constantly and likewise unstable landscapes and backgrounds. Looking forward to try it locally either way as I'm not a big fan of mega corp cloud stuff or discord bots to say it in diplomatic way.
yeah a video to how install it localy would be great =)
Is there any opensource model that can do video to video similarly to gen2?
is there any open source that can generate like more then 15 second of video
Would love a video about how to install this locally, like you mentioned. Thank you.
Remember, this is only 2023... imagine what this will look like in 2467!!!
Does this involve me going to github and filling up my hardrive with garbage that doesn't work and just throws off a million error messages? Because I'm already pretty tired of that stuff
5:33 smooth transition!
What are ur thoughts on kaiber
Nvidia sent you a RTX 4080! WOW! Awesome! I guess they consider you promoting them, since we want to do things that require an Nvidia GPU.
My only reason to upgrade to a newer Nvidia GPU, is for faster AI generation, which means I'll need the best of the best, so wise move on sponsoring you!
Why are the videos so short? Can they be longer?
Can we run this on stable diffusion?
The next will be and ai DAW vst au voice generator where you can imitate any voice or take vocal recordings to a higher quality, the maker of this will be the next level
I grew up in Ft Myers 😂 nice shirt 👕
Love your videos! Local install demo, yes please
Please show us how to install locally
This is a game changer.
👇 Yes?
Can you do a tutorial on how to train this? video length etc
Imagine someone making ARG’s out of AI Videos that will be cool 😅
so far, the outputs don't come near as close to what I want as Gen-2
The real dancing lemon character was the friends we met along the way.
Can you run it or install it locally?
can you make a tutorial how to get access to the run way ml discord server, i really want to try this stuff out
Guys is there a way i can find the name of that *satisfying car game* used to boost screentime in short videos and tik tok videos?
Any img search method you know?
Satisfying car game is vague
gen2 was released to the public today
I'm sure under the hood it's cool, and in a year or so it will go from Potat0, to Potat5head, at which time it will be cool. I think this current stuff is pretty unimpressive in terms of the other AI stuff going on. But hey, thanks for covering this.
5:25 TH-cam doesnt like this
"with the midjourney the dalle .. the bing.." matt turned into bill cosby
MattVidPro, I'm your biggest fan!
I dunno, I've been getting some really coherent high res video from Gen 2 lately, been playing with the CFG scale a ton, and a clear and concise prompts with the right keywords usually gets you there. Adding a driving image helps a lot too, but lately, I've been getting sick results without the need of an input image :)
Actually the free versions not really gonna work for most, as it runs out of memory pretty quick, even on a 16gb card. This one's definitely gonna need 20 gb or 24 gb
All are impressive.
yeah, i'm very lucky, this is gonna be blazing fast on my gtx 1050 2gb laptop!
Matt you tend to waffle on a bit - you should have segments on the red TH-cam line so we can see each part of the video and go to it - like setting up in collab.
If you post the video of you doing this with your graphics card. I'll test it out with a couple of 3090's.
You know that HD sizes are for the height right?!
So this means a minimum of 720px to be HD... This is not HD though!
1:28 Who else thought they got a message?
Best part are the shutterstock watermarks.
tell google collab to stop banning our research!
Love it
Thanks 😊
Welcome 😊
the videos that are "better" then the standard old model are the ones from the standard not new model 6:45 epic fail
Can't wait to take my favorite books and have them made into movies 🧠
That's really cool if that's continuous GPU for free. Hehe.
Brilliant as always 🔥🔥💯💯
Thanks 🔥
@@MattVidPro 🇹🇲
waiting for blender tuto :)
Wow!
Trained model w Shutterstock images 😂🤦🏾♂️
great
I hate to keep bringing this up to countless youtubers, but it's honestly a pet peeve of mine. Please mute your discord when recording videos, it's not only annoying that it's in so many youtube videos now, but it's also confusing for people who have discord as well and think they're being messaged. Not to be a dick but just a humble suggestion.
Julian Casablanca became an youtuber...
Dayum
do the local computer install buddy
Pretty sure it's PotAt, not "Poteit"
please use ai to add a curtain to your window
I need blackout shades
14h 912 likes 197k subs
Open source is king
open source rule, soon people will fine tune it
Oh yea. I have stopped chatting with gpt4 just to watch the video and I guess that I will return doing that after these 16 minutes hahah xD
With 25 messages per 3 hours?
@@omnigeddon Whenever I can. And when something tackles my mind I can also do some research on my own. We are sort of cooperating. Nothing too special due to limitations you have just mentioned, but trying to test as much as I can.
RIP Gen-2
Your shooting style screams for arri alexa 35 if you don't want to set up extra lights and put ND on your windows 🤣
Upgrading soon
what has the world come to?
ai
Our society.
it's idiotic these ai developers won't do windows installers instead of us f ing around with github, phython
😅
local guide!
Bruh im from ft myers
My friend gave me the shirt lol
Make no mistake, aint nothing beating midjourney in terms of quality. All newcomers continue to fail in comparison...midjourney is lightning in a bottle...
I would of checked out Yet WATER marked no Thanks! Not worth my time.
Dude you talk too much !!!! get to the point , how to use it or download it or anything ... i play on 2x playback speed and still got bored
Dear NVIDIA, I am poor boy, please send free 40x series GPU, I will change world!
Sincere
I'm not gonna lie, you're REALLY over hyping this. In terms of doherency, from what I've seen Gen2 is better.
please make videos on installing this on blender or local? let us know is there is a sign up or when the new video will launch ;)