check out Storyteller AI at www.storyteller.ai/?ref=bycloud and join their discord now! discord.gg/storyteller this new age of internet is about to be chaotic 💀
Working for an big AI company right now must be extreemly stressful. Months of work wasted and feels like nothing when some other company comes with something better right after you. They all trying to be the first, the best while everything have to be pushed out unfinished this way to compete with the others like the smartphone market with a lot of unfinished products. But pushing it out without a release made all the other companies trying to do the same but better while OpenAI didn't released it yet.
It’s hard to say for sure, depending on their real intentions. They’ve said that at times they want to warn people about what is coming to give them some time to adjust. But of course it also ends up making everything progress faster than it probably would have if they hadn’t demoed it to begin with lol.
@@Amesioses Don't get your hopes too high up, that's not going to happen. AI has not progressed linearly through history and there is no reason to think that it will in the future.
@@Hiestaa AI models will be able to emulate games in real time within the next 14-20 months. After that, there will not even be a need to develop games.
Wait... is “Toys R Us” supposed to be a play on words? I know it's not pronounced "Toy Saurus" but I never realized the name could be pronounced in such a dinosauric manner!
I always assumed that Sora was trained on a nerf or gauss splat method in the pipeline bc a lot of it's output had that 3d photorealism look It seems obvious to me that other models are doing this too It makes sense bc it gives you an easy way to auto lable synthetic data
I think it's Unreal Engine footage as someone else guessed too. And photoscans actually are often used in Unreal Engine for the realism while Unreal Engine it'self provides a lot of stuff for the physics and understanding of the scene.
@@MrGTAmodsgerman Well that does not really change if assets were generated from a nerf or guass splat...you can port that to unreal. The point is I think a lot of the assets vid gen models are trained on are not just native raw irl vid footage
@@memegazer Synthetic data usage is ofc a common method in AI training. But i don't see why it have to be nerfs. Nerfs don't provide that crisp quality and it's still in a quite raw state, guassian splatting also. And also very limited for now. It can help to mimic real lighting in scenes but mostly i think Unreal is for (if it was used), is the physics and simple process of segmentations. They don't have to do segmentations with other AI on the rendered videos, they could export it directly with it.
@@MrGTAmodsgerman I just noticed the photorealism look and nerf or gauss splat would be a much quicker way to make assets than hand crafting them. It seems obvious to me that 3d software was involved in the training...and that makes sense bc it is easier to lable key data points like camera and lighting, that raw video scraping probably won't have until you can train a labling model with a 3d pipeline to process raw video
@@memegazerI do also seem to remember some people pointing out some of the Sora footage having artifacts at the edges that seemed like splats as well (those kind of floating specs).
Not just Genshin. The Touhou fans will have a field day too. They would use MikuMikuDance to have a 3d base and then use the AI to add the video filter on top.
Yeah there was a video a while back (maybe a year ago?) from an artist who was much more balanced than most on the topic of AI images. He pointed out that part of what was so interesting about them then was the roughness and strangeness of the outputs. He felt like we’d lose things as it got better and I think it is true. Hopefully we’ll have people that keep using older tools or purposefully train ones that aren’t as “good” for some retro AI flavor. Like the people that make pixel art or chiptunes.
At this rate creativity is gonna die down quite fast , all the stuff that we are seeing would eventually become recycled creativity of some ancient erra.
@@bycloudAI Well that's true but seeing how RUclips has approximately 300k hours of content(googled number) being uploaded daily , the quantity is not the problem. It would be nice if the quality of these (AI)videos could increase too. And by quality I don't mean visuals , I mean the artistic intent behind those vids.
@@memegazer Well , I don't mean it like that. The thing is , AI can only be trained on existing data , it cannot think or be creative like humans as of now. The best it can do is take the already existing content and mix it up accordingly to generate something "new".No doubt it is helping researchers tremendously in some fields but I was just talking about the videos or imagery that AI produces online and how it will form an era of no/minimal artistic creativity. P.s. sorry for any grammatical error.
these productss are way too expensive given how primitve the tech is and how often the generations come out like dog shit, paying for 150 a month means you might get 2 or 3 videos you like
I love the artifacts in AI video. I mean, it's really cool, don't get me wrong but only dumb, half blind, people would be fooled. I mean who eats food that appears out of their mouth as they take a bite and then their hand melts into the food etc... anyone paying attention is not fooled, even on these low res, grainy clips of 2-3 seconds you can call them out immediately. ML has been around for decades, I honestly don't think we're going to see an explosion of AI's ability but rather an explosion of people trying to capitalise on the hype and a gradual improvement albeit a long way off it really being of any use without significant human intervention to pave over the cracks.
Awesome video all in all, I have wanted to know a recap of what's happening in the last few months in text to video and this answered everything I had in mind! :) Thanks. ❤️
well i dont pay for the new model cause they're just to costly for me and alsoo after 1-2 weeks some other company makes something more good and more cheap and also some for free and open source
I'm not surprised that this is happening, I'm surprised how fast it is happening. I'm betting that we will have a full length AI movie in the next .. 16 months
I agree that we're moving faster than expected, and I hope we only go faster from here. Somewhere in January 2026 we might get something based on Christmas/New Years 2025 that will break the 1 hour mark in 720p, now that would be wild.
No, they are not. There is a long way before they will become useful. Unrealistic scenes can sometimes look decent but as soon as you try to create something from this world it's 80% artifacts and 20% of ok content.
"Can have people holding chopsticks without crazy deforming" - literally watch the 'woman' in the middle's chopsticks... looks like a pool noodle going through a wormhole ffs.
You said it very accurately that AI needs to get some decent reasoning capabilities. This is my biggest gripe about both image and video generators. For any human it is dead easy to tell if a person has all 5 fingers while holding a mug when some digits may not be visible. We know that a body could never twist and bend and morph in real life, but this is very common in AI generated visuals. We really need to train these models by teaching them some actual physics and reasoning so we can control when and if we want to keep or break realistic physics.
Hi bycloud, the moment I saw your video thumbnail shows up in my feed, I decided to subscribe to your channel so that I can comment on your new video as soon as possible, stating that I am going to block your channel and those shameless plagiarizing thumbnails from my feed forever. You must believe your videos are so good that it deserve to be seem by more people than it was, so your clickbait thumbnails are justified. But no matter how good (you think) your video contents are, the moment you pull these dirty tricks you lose all respects and views from people like me. Hope you come around one day but I'll never know anyway.
in LLMs, there smth called self-play preference optimization, which uses a small preference dataset as the reward model for closed-loop optimization of the LLM so yeah, totally possible
No idea about cling, but have used Luma dream machine and Runway gen3, and the outputs are unusable, it doesn’t do what the prompt says and every single video has comical morphing. They probably produce thousands of images and show the 5 or 6 which turn out sane. Other than that I tried using video generators locally, I have a decent beefy system, nothing spectacular but not a regular system either, and any model simply runs out of memory. Overall I feel it is less hassle to just shoot/render these videos instead of haggling with a model. ChatGPT is great, but it’s been 2 years and even image generators have more or less remained the same, and I am not expecting video generators to become better anytime soon, or may be ever.
Every time i see these thumbnails, i think fireship has released a new video, and then i notice it's just one of the garbage bootlegs lmao. Good thing there's the "never recommend this channel to me" button on youtube.
Well we can't tell if you are a new kind of bot that claims to be the first non-bot. 😂 But since I see other comments being 1-2 minutes faster I actually believe you. 🎉
In a few years, a group of people will make it a hobby to distinguish between what is AI generated and what's not. Just like how we are understanding what's an AI generated image right now. It doesn't matter, AI generated content will always have a bottleneck.
this all looks like shit to me, a year ago i would have said AI video generation is 10 years away, with these new releases im going to say AI video generation is more like 20 years away.
@@gfdggdfgdgf either your comment, or the one you replied to will feel very out of touch in 10-20 years. My guess, as some random guy who claims to have some insights into this field, is that it'll be yours. Now go search for other similar voices, they aren't very loud because the ones going against the hype usually get tuned down, but they are here, and they are becoming louder by the day. AI Winter is coming. And I am looking forward to it so machine learning engineers can start building useful AI products again.
@@Hiestaa I posted this last year and still stand by it. It's coming whether we want it or not. In the not too distant future the media experience will completely change. AI is progressing at a crazy pace. We're much closer to no actors needed than anyone would have expected a few years ago. There could be advantages as well, we could have new period accurate movies starring James Dean, Cary Grant, Marilyn Monroe, Audrey Hepburn etc. Biopics starring the actual (generated) person, terminator 1 or 2 with Stallone, Indiana Jones with Tom selleck, the possibilities are endless. It's not unlikely that eventually you type in something like play Casablanca but in color with al Pacino playing the main character and the movie will play. Sounds far fetched? To most people yes but eventually it will happen. Change weather conditions, time of day, actors, anything in real time. At one point there will be a cut off point, anything up to that point is "real" media, anything after that a high chance it was generated. How this will evolve is anyone's guess, personally I feel it will start with realtime colorization of movies, realtime changing of aspect ratio, then replacing actors, then environments, then change scripts/stories and eventually movies from scratch.
how long do you need to make a video that really confuses someone else than my dead grandma? how many you attempts required until its not a nightmare fuel?
If you think about how hard the AI community was working 2-3 years back just to produce blurry image that might look like a dog... it's enough to freak out.
There are AI models who identify that they are AI created models and people still think they are real! No company should be paying human models for anything 😂
Its sort of crazy to think that there are a lot of quite good AIs for VIDEO generation... But for text generation only openai and anthropic are the good ones.
man I was so close to showing this to my team at (global enterprise) before you had a sponsor section. Don't get me wrong, I love you getting the monetary value you deserve, but hey.
check out Storyteller AI at www.storyteller.ai/?ref=bycloud and join their discord now! discord.gg/storyteller
this new age of internet is about to be chaotic 💀
Openai was so dumb to show their cards when not ready to play
A lot of companies do that. Usually it means their product isn't actually THAT good. Or they don't want the backlash if it gets "misused".
Working for an big AI company right now must be extreemly stressful. Months of work wasted and feels like nothing when some other company comes with something better right after you. They all trying to be the first, the best while everything have to be pushed out unfinished this way to compete with the others like the smartphone market with a lot of unfinished products. But pushing it out without a release made all the other companies trying to do the same but better while OpenAI didn't released it yet.
Well now everyone knows of Sora. that's a win on their end
It’s hard to say for sure, depending on their real intentions. They’ve said that at times they want to warn people about what is coming to give them some time to adjust. But of course it also ends up making everything progress faster than it probably would have if they hadn’t demoed it to begin with lol.
They only want to be relevant for inversors. It's never about the users
Not the distracted boyfriend into rickroll💀
We've come a long way from Will Smith eating spaghetti.
It was only one year gap , now after 2-5 years , we will make our own video games and movies and books
@@Amesioses Don't get your hopes too high up, that's not going to happen. AI has not progressed linearly through history and there is no reason to think that it will in the future.
@@HiestaaThat’s actually great news
@@HiestaaYou’re right it’s grown exponentially. That means his idea is probably closer than you think. Don’t be a party pooper lol
@@Hiestaa AI models will be able to emulate games in real time within the next 14-20 months. After that, there will not even be a need to develop games.
looks like ASMR videos are gonna get AI generated in 2 years from now on
😅 I prefer pre recorded asmr vídeos
Excellent
@@Pawnsappsee I don't
@@lcmiracle Bro doesnt like life
@@lcmiracle do you have any examples of AI gen ASMR?
Thanks for the shoutout, Bycloud! And fantastic rundown here of the latest and greatest in AI Video!!
Bro did you just make me watch a Story teller AI sponsor without me realising it?!
that segue was way too smooth frfr
It helps that it does actually seem pretty cool looking…
Made me actually want to try it too, that was the best sponsorship segue I've ever seen.
You know it's good when you don't even notice it. It was hard to notice. His last line in that segue gave it away.
Toysaurus... might need to train your text-to-speech ai a bit.
yikes I think my TTS didn't grew up in NA
Wait... is “Toys R Us” supposed to be a play on words? I know it's not pronounced "Toy Saurus" but I never realized the name could be pronounced in such a dinosauric manner!
tbh i always thought that's what the name stood for, i'm not us based though
This is the worst it will ever be.
I always assumed that Sora was trained on a nerf or gauss splat method in the pipeline bc a lot of it's output had that 3d photorealism look
It seems obvious to me that other models are doing this too
It makes sense bc it gives you an easy way to auto lable synthetic data
I think it's Unreal Engine footage as someone else guessed too. And photoscans actually are often used in Unreal Engine for the realism while Unreal Engine it'self provides a lot of stuff for the physics and understanding of the scene.
@@MrGTAmodsgerman
Well that does not really change if assets were generated from a nerf or guass splat...you can port that to unreal.
The point is I think a lot of the assets vid gen models are trained on are not just native raw irl vid footage
@@memegazer Synthetic data usage is ofc a common method in AI training. But i don't see why it have to be nerfs. Nerfs don't provide that crisp quality and it's still in a quite raw state, guassian splatting also. And also very limited for now. It can help to mimic real lighting in scenes but mostly i think Unreal is for (if it was used), is the physics and simple process of segmentations. They don't have to do segmentations with other AI on the rendered videos, they could export it directly with it.
@@MrGTAmodsgerman
I just noticed the photorealism look and nerf or gauss splat would be a much quicker way to make assets than hand crafting them.
It seems obvious to me that 3d software was involved in the training...and that makes sense bc it is easier to lable key data points like camera and lighting, that raw video scraping probably won't have until you can train a labling model with a 3d pipeline to process raw video
@@memegazerI do also seem to remember some people pointing out some of the Sora footage having artifacts at the edges that seemed like splats as well (those kind of floating specs).
3:05 genshin people are gonna have fun with that 💀
Not just Genshin. The Touhou fans will have a field day too. They would use MikuMikuDance to have a 3d base and then use the AI to add the video filter on top.
the worst thing that happens to these ai generators is that they get better. I can't get enough of the weird , surreal stuff.
Yeah there was a video a while back (maybe a year ago?) from an artist who was much more balanced than most on the topic of AI images. He pointed out that part of what was so interesting about them then was the roughness and strangeness of the outputs. He felt like we’d lose things as it got better and I think it is true. Hopefully we’ll have people that keep using older tools or purposefully train ones that aren’t as “good” for some retro AI flavor. Like the people that make pixel art or chiptunes.
Toysaurus 🦖
At this rate creativity is gonna die down quite fast , all the stuff that we are seeing would eventually become recycled creativity of some ancient erra.
Another aspect is that we are reaching a point where there will be infinitely more content than we can humanly consume.
If you are going to look at this way then the acient quote "there is nothing new under the sun" summed this idea up a long time ago
lmao capitalism already did that. We've been recycling and selling shit back to people for 50 years.
@@bycloudAI Well that's true but seeing how RUclips has approximately 300k hours of content(googled number) being uploaded daily , the quantity is not the problem. It would be nice if the quality of these (AI)videos could increase too. And by quality I don't mean visuals , I mean the artistic intent behind those vids.
@@memegazer Well , I don't mean it like that. The thing is , AI can only be trained on existing data , it cannot think or be creative like humans as of now. The best it can do is take the already existing content and mix it up accordingly to generate something "new".No doubt it is helping researchers tremendously in some fields but I was just talking about the videos or imagery that AI produces online and how it will form an era of no/minimal artistic creativity.
P.s. sorry for any grammatical error.
Another pro of Mukbangs is that they made a mobile app instead of an awkward website like our video generators
these productss are way too expensive given how primitve the tech is and how often the generations come out like dog shit, paying for 150 a month means you might get 2 or 3 videos you like
For you. Resource supply vs demand
Its probably because it takes millions to produce each model, they probably wanna make a ROI way before the next model so that's what happens.
Yeah, not worth it right now, for sure. Eventually it might, though.
The drew gooden vid was a banger
I love the artifacts in AI video. I mean, it's really cool, don't get me wrong but only dumb, half blind, people would be fooled. I mean who eats food that appears out of their mouth as they take a bite and then their hand melts into the food etc... anyone paying attention is not fooled, even on these low res, grainy clips of 2-3 seconds you can call them out immediately.
ML has been around for decades, I honestly don't think we're going to see an explosion of AI's ability but rather an explosion of people trying to capitalise on the hype and a gradual improvement albeit a long way off it really being of any use without significant human intervention to pave over the cracks.
2:20 ZZZ shown!
3:05 Genshin shown!
ZZZ!
Awesome video all in all, I have wanted to know a recap of what's happening in the last few months in text to video and this answered everything I had in mind! :) Thanks. ❤️
4:53 lol is this the official mascot of stable diffusion 3 now
well i dont pay for the new model cause they're just to costly for me and alsoo after 1-2 weeks some other company makes something more good and more cheap and also some for free and open source
I'm not surprised that this is happening, I'm surprised how fast it is happening. I'm betting that we will have a full length AI movie in the next .. 16 months
I agree that we're moving faster than expected, and I hope we only go faster from here. Somewhere in January 2026 we might get something based on Christmas/New Years 2025 that will break the 1 hour mark in 720p, now that would be wild.
Those waving hands is the storyteller video were so cringe it’s hard to believe it’s from their own marketing material 😅😅🔥
Why you link the gooden video (which is already being spammed everywhere ) and not all the shout-outs you mentioned?
Bro why you say ToysRUs like that?
Good video bro
No, they are not. There is a long way before they will become useful. Unrealistic scenes can sometimes look decent but as soon as you try to create something from this world it's 80% artifacts and 20% of ok content.
Man, this guy believes any video from AI developers that are made to bait investors
"Can have people holding chopsticks without crazy deforming" - literally watch the 'woman' in the middle's chopsticks... looks like a pool noodle going through a wormhole ffs.
thumbnail at 5:37
Woah the storyteller thing is insane, I was talking about a tool exactly like that 4 years ago and it's actually happening
AI will kill the Internet!
You said it very accurately that AI needs to get some decent reasoning capabilities. This is my biggest gripe about both image and video generators. For any human it is dead easy to tell if a person has all 5 fingers while holding a mug when some digits may not be visible. We know that a body could never twist and bend and morph in real life, but this is very common in AI generated visuals. We really need to train these models by teaching them some actual physics and reasoning so we can control when and if we want to keep or break realistic physics.
Hi bycloud, the moment I saw your video thumbnail shows up in my feed, I decided to subscribe to your channel so that I can comment on your new video as soon as possible, stating that I am going to block your channel and those shameless plagiarizing thumbnails from my feed forever. You must believe your videos are so good that it deserve to be seem by more people than it was, so your clickbait thumbnails are justified. But no matter how good (you think) your video contents are, the moment you pull these dirty tricks you lose all respects and views from people like me. Hope you come around one day but I'll never know anyway.
If you could measure sexual arousal you could make a closed loop system with these AI video generators to maximize arousal.
in LLMs, there smth called self-play preference optimization, which uses a small preference dataset as the reward model for closed-loop optimization of the LLM
so yeah, totally possible
WTF? You are not code report....
After one year (ie, July 2025) we might be talking about how to distinguish between completely AI generated videos vs generic ones 💀
11:26 WTF LOL
anybody else hungry?
I knew back then we were talking about upscaling AI.
Now we are talking about AI videos and audio lol..
toy saurus lol
No idea about cling, but have used Luma dream machine and Runway gen3, and the outputs are unusable, it doesn’t do what the prompt says and every single video has comical morphing. They probably produce thousands of images and show the 5 or 6 which turn out sane. Other than that I tried using video generators locally, I have a decent beefy system, nothing spectacular but not a regular system either, and any model simply runs out of memory. Overall I feel it is less hassle to just shoot/render these videos instead of haggling with a model. ChatGPT is great, but it’s been 2 years and even image generators have more or less remained the same, and I am not expecting video generators to become better anytime soon, or may be ever.
At the moment content is wierd.
Ai corn when?
Every time i see these thumbnails, i think fireship has released a new video, and then i notice it's just one of the garbage bootlegs lmao.
Good thing there's the "never recommend this channel to me" button on youtube.
I did not know about the "never recommend" button.....
I love your videos.
And your sarcasm and comedy taste is sooooo good
First... non-bot comment?
Well we can't tell if you are a new kind of bot that claims to be the first non-bot. 😂
But since I see other comments being 1-2 minutes faster I actually believe you. 🎉
In a few years, a group of people will make it a hobby to distinguish between what is AI generated and what's not. Just like how we are understanding what's an AI generated image right now. It doesn't matter, AI generated content will always have a bottleneck.
😅 do you know people are also mistaken artist art for ai? I don't think that will work well.
This will not age well. A(G)I will have less of a bottleneck than humans do in creating anything.
2:11 ayyy! Vancouver represent!
Next generation of porn is going to be out of this world.
Well done as always 👍
how do they do the smooth transition from another video to rick roll with ai?
Ai getting out of hand
You're getting out of hand! 😂 Jk 🕊️
Wait till you see the female robots.
@@NoidoDev I'm sick of waiting
@@digestiveissue7710 Yup, need the goth mommy robot NOW!
wtf
“Not too wrong” was the same phrase I used after watching that Drew Good video too. Glad to see that we agree there.
Edit: Also, great video!
Openai just posted a video of Sora now
?
this all looks like shit to me, a year ago i would have said AI video generation is 10 years away, with these new releases im going to say AI video generation is more like 20 years away.
This comment will not age well
Bait used to be believable
@@gfdggdfgdgf either your comment, or the one you replied to will feel very out of touch in 10-20 years. My guess, as some random guy who claims to have some insights into this field, is that it'll be yours. Now go search for other similar voices, they aren't very loud because the ones going against the hype usually get tuned down, but they are here, and they are becoming louder by the day.
AI Winter is coming. And I am looking forward to it so machine learning engineers can start building useful AI products again.
@@Hiestaa I posted this last year and still stand by it. It's coming whether we want it or not.
In the not too distant future the media experience will completely change.
AI is progressing at a crazy pace. We're much closer to no actors needed than anyone would have expected a few years ago. There could be advantages as well, we could have new period accurate movies starring James Dean, Cary Grant, Marilyn Monroe, Audrey Hepburn etc. Biopics starring the actual (generated) person, terminator 1 or 2 with Stallone, Indiana Jones with Tom selleck, the possibilities are endless.
It's not unlikely that eventually you type in something like play Casablanca but in color with al Pacino playing the main character and the movie will play.
Sounds far fetched? To most people yes but eventually it will happen. Change weather conditions, time of day, actors, anything in real time. At one point there will be a cut off point, anything up to that point is "real" media, anything after that a high chance it was generated.
How this will evolve is anyone's guess, personally I feel it will start with realtime colorization of movies, realtime changing of aspect ratio, then replacing actors, then environments, then change scripts/stories and eventually movies from scratch.
Yeah, you are probably right. How do you know these things? You must be extremely tech savvy.
7:45 the stalin photo was both hilarious and sad :/
Great video. There a lot of... Eating. 🥑
Clickbait title....a video used to promote ai service to make even more ai video.
how long do you need to make a video that really confuses someone else than my dead grandma? how many you attempts required until its not a nightmare fuel?
the interrupted by ai is one of my favourite tik tok trends
Sora lost before it even started
when YOU say that, you know it's BAD
Eventually this will be society disrupting technology and there's no way to stop it anymore.
Nice
Imagine when it decides to reveal the secrets of the universe
Its inspired from the Animated Diff Community, and not really inspired, but totally 'borrowed'.
This is actually wow! This is shocking if true!
If you think about how hard the AI community was working 2-3 years back just to produce blurry image that might look like a dog... it's enough to freak out.
cool
There are AI models who identify that they are AI created models and people still think they are real! No company should be paying human models for anything 😂
your video has a bland enough script that it could be ai generated...
King is substantially better than runway 3 and Luma
just turn of the nsfw filter and people will perfect the prompting lol
Stop. Generative. AI.
No
Too late.
Stop image generation but i think realism video generative AI & text based AI should stay.
why
❤❤❤❤❤❤❤
Its sort of crazy to think that there are a lot of quite good AIs for VIDEO generation... But for text generation only openai and anthropic are the good ones.
Tbh we got LLAMA3 as well. Phi-3 is also great considering its size.
stop simping for ai, what a joke channel
sounds like you don't get the joke. :P
whats wrong with ai?
Better get used to it pal
man I was so close to showing this to my team at (global enterprise) before you had a sponsor section. Don't get me wrong, I love you getting the monetary value you deserve, but hey.