Just a quick compilation of all the User Generated Sora videos that were recently "leaked" on Hugging Face. I tracked (I think) all of them down, and was able to provide the prompts used. It's kind of a fascinating look at what Sora can and cannot do. No Commentary on this one, just the videos!
@@LouisGedo Thanks! This was an old Ambient experiment from back when I was doing more music stuff on the channel. It was a lot of fun to make-- not AI! Fun with Synths!! I can get lost doing this kind of stuff for hours!
There is a new update in google imagen 3, where now you can create images with several aspect ratio options. Maybe it can be an idea for your video content 👍
@@arimuraminami I'll check into Imagen again. I took a look at 3 when it first launched and I wasn't overly impressed. But, that happens a lot at launch. The models always improve.
Good one! That and the fact the OPENAI is paying Africans peanuts and giving them PTSD: ua-cam.com/video/qZS50KXjAX0/v-deo.htmlsi=C0ck4SARUPix3LK5 as they detailed in 60 Minutes.
I think what's fascinating about this is when you examine the prompts. I think Sora needs a lot more prompt help than we think it does. We've been getting used to LLMs as an in-betweener with those "magic prompt" enhancers. Clearly, Sora is in need of that. If only OpenAI had access to a LLM...hmmmm...
Is it possible that the lack of coherence is due to the artist not having enough GPU power at their disposal? Maybe these weren't rendered on the OpenAI servers...
I was thinking the same. We've seen hundreds of examples better than the best stuff presented here. Open AI just kicked off the revolution with the presentation of Sora and everyone has been breaking their necks to catch up. And well, they've caught up alright. Real competition is in play. May the best models win.
I've been saying this for months. The reason they won't release sora is because once they do the grift is up. The examples we saw at the beginning of the year were cherry picked and the thing is a slop machine. Things people actually want to watch are never going to be created by text prompts
5 днів тому+5
3 video updates in a row, you are on fire this week 🔥
Haha, this last one was kind of spur of the moment. Onward to Thanksgiving and off the rest of the week! No matter WHAT happens in AI! (famous last words...)
I applaud Sora for raising the proverbial AI video bar. It’s still closed vaporware so far as I’m concerned, until the public can get its dirty hands on it. Hopefully this changes soon…. Great video 🍻
Agreed-- but, I kinda missed the boat on this one (funny enough, I was getting a new dishwasher that day), and I kind of feel like all the "hot takes" made the rounds. But, I've got a lot of thoughts on it, and saving them up for the next big Sora news cycle story. Hopefully, that will be the actual release? Sigh...
Man, I kept thinking the same thing! You have one magic Sora prompt, and…that’s what you went with? Interesting. I shudder to think what would happen if 3-Wish Genies actually existed. 😂
Sora reminds me of that simpons scene where mo steals homers idea and homer is in mos now packed bar screaming "you just lost a customer" . Mo is shouting back 'i cant here you homer" as the till rings out with the hundreds of punters buying beer.
Awesome to hear! Yeah, this was exactly the video I thought we needed. Lots of commentary videos, but I really just wanted to put one up that just showed the footage. No memes, jokes, or blathering. Just the footage. And, y'know-- obviously we can chat about what we think in the comments!
What's the point? The train is going faster every day and SORA is still at the station, reading the timetable. Among the untouchable models the Meta Movie Gen looks far superior to this and with sound.
With you on that. Funny enough, months before Meta dropped their announcement, the Sora team did an interview with MKBHD where they mentioned sound for Sora 2. But...typical OAI, they missed the kill shot on the announcement.
Nice to see real videos, not only the top ones they show when try to sell the product. I hope this makes they be more rumble at pricing. There are glitchs and fails as in all other models.
Here's the good news: some content was generated without being flagged and blocked! Thanks for the no-comment video and prompts so we can try to absorb it all. I see some limitations, but also more usable content, especially dynamic motion, albeit slow-paced. Wondering about how long it takes to generate content and any cost compared to others charging $95. Hopefully, you can get an official eval!
100%! I figured all the other channels had done their hot takes, but no one actually comp’d everything together. Haha, I did miss my chance for the “hot take” because I had to get a new dishwasher. Haha, glamorous life of UA-cam! So, I figured this would serve as a good little time capsule for the moment we got to play with Sora. For sure there are some “eh” outputs, but there’s also a lot of usable stuff in here. And since I was able to track down the prompts, I think it serves as a really fascinating look at what you can and can’t get out of Sora.
😂 last part. Yep noted but appreciated your comp because of all the hot take noise elsewhere. Will look forward to yours none-the-less. Happy Thanksgiving!
Ha! Thank you! I meant to run them through a translator, but-- then I would have missed your comment! haha, appreciate it! And that's also, pretty funny!
Haha, that's me-- but, kind of aping Phillip! And NOT AI as well. I actually broke out a bunch of synths a year or two ago for a little mini EP experiment. It was a lot of fun!
Thanks for the showcase. Some of these are excellent, some not so much but I also believe it isn't the full model in use. The thing is, even if Sora does release in some form, there's more competition now than when the earlier footage was seen and it's only going to increase. Question: Now you've seen this footage, do you think that earlier footage was genuine, aside from being heavily cherry picked? I seem to recall a lot of folks claiming it was fake.
@@sidbrun_ Who knows? You can't deny that the current evolution of AI is very fast, like it or not. I'm a musician and visual artist, and I say armbrace the new or die.
@@-Deena. Good mindset you have. I don't even have to wish you the best in this AI world because you've already taken the steps to succeed in this new reality. (which we saw coming from a mile away)
@@jojothepro15 Thank you. We all have to remember that at one point oil paint, photography, film, were considered the work of the devil to some, and unless practicing artists embrace new technology, AI will end up producing only clichéd teenage nerd boys sexual fantasies, and no one wants to see that!
Some time ago, I watched a video where you compared Sora as the generative AI video standard to other video creation companies. I mentioned that it felt unfair to include Sora in the comparison because it’s not even available yet. You responded by saying it was like refusing to rate an ARRI Alexa against other cameras simply because we don’t have one in our possession. My point wasn’t just about not having access to Sora-it was that none of us have been able to thoroughly test it like we have with other platforms like Kling, Runway, Hailou, Kaiber, and others. With those, we’ve had the opportunity to explore both their strengths and areas for improvement. In contrast, OpenAI keeps receiving credit for Sora’s potential without even actively participating in the competition. And now they have, and they are lacking.
Generator? Currently I think the crowd favorite is Minimax. Unofficial, of course-- but that's just the one that I've noticed people seem to really like the most.
Agreed. The smoothness of motion in characters is fairly stunning. Although, I have to admit, personally I find the T2V outputs to be a bit bland. Image to Video? That's going to be stunning.
Don't forget runway with midjourney sora can't compete with no one yet as it's is not available to the public runway is better as its 80 a month for unlimited minimax and kling don't offer unlimited
I want Sora's results with runway's render time :)) we still need to figure out how long was the waiting time, kling for example, have gone up from 7 minutes to 13 minutes for PRO users, that's sucks, and highly inefficient, if Sora can bring Runway's render time, they got me!! as for the overall video quality, it looks like the model was mainly trained on National geography channel. the animals are superb, the people,,,, hmm... not so much... but then again, just a few months ago 70% of the people in the comments would argue that it's not real, and open AI are faking "ai generated videos"... so we are in a good place :)
Imagine that you're in a light sleep, between dreams, just barely able to hear the sounds in the room around you, and someone says something -- describes a scene or event -- and you start dreaming about it. That's a prompt. And that's often about as good as Sora's prompt adherence is. What we need is something that reliably follows prompts, in their entirety, and understands the meaning of the prompt to the degree necessary to accurately follow it. Sora isn't there yet. As such it's limited in its utility for things like movie making. It's hit or miss. Given how expensive it is, how much energy is consumed, how much cooling is required, the compute resources needed, we really can't afford a technology where it produces a bunch of unuseable results for every one that does actually adhere to the prompt with sufficient accuracy. They need to fix this before Hollywood is doomed. I trust they will succeed, or someone else will, and only then will Hollywood be doomed -- or at least will have to adapt its processes and techniques.
10:42 That was really funny 🤣 12:14 I thought he would throw the ball 13:38 The prompt was literally "blue sky", and it really was blue sky. dude missed the opportunity to make a cool prompt 🤷♂
Fully agree. If this ISN'T a PR stunt, because I'm so turned around, there's a 35% part of me that thinks it might be-- then yeah, this was a gold mine for them.
Just curious: Did you upscale these? The twitter versions did not look near as good…😮 And I think the Sora turbo model understands anime a lot better than the competition. Imagine what the prime will be capable of? Also, loved the music!
@@TheoreticallyMedia Yeah, I know what you mean. All this AI and the energy it takes to keep up with it has taken its toll on me too. I am a US Vet (in the same age range) who back in the day used to record all my music by myself and mix/master, do all the release legwork and artwork, etc. (no AI of course!) and now somehow I can't find the time to pick up my old Les Paul anymore...Lmao! 🥲
Honestly, I don't think SORA is better than Runway, Kling, Hailuo, or the like. It would be great if Open AI can make SORA more affordable than others.
most people doing pic to vid now days much more control but still the jank if you look hard enough so I wonder how well it handles with pics/vids hopefully it has video to video
I think it needs a LOT more coaxing from the prompt. Simple prompts (for now) aren't cutting it w/ Sora. Hmmm, if only OAI had a LLM to help. Shame they don't...haha.
Nice music!!! Made watching these really feel atmospheric. Have to say tho... Im not terribly impressed with the Sora generations. A couple were great... particularly some of the anime... but really Im seeing a lot better from other generators. and yes i do know its Sora Turbo... but still.
I told everyone that if they thought a company that was miles ahead has just been twiddling their thumbs for a year then they would all be proven wrong. And well, yup. Especially if this is just turbo. A lot of these look a year old as well - the proper Sora 3 is gonna be amazing - this also shows that SORA is the most flexible model in terms of different topics it can do where as kling minimax etc excel at certain things and then completely flop on others.
In Feb when we first saw Tokyo Woman, if you told me we (still) wouldn't have Sora by then...I'd sadly nod. But if you told me we'd have 20 other versions that were just as good-- and in some cases better? I'd be STOKED. It's not even December yet, and I'm really happy with where we are.
@@jzwadlo Ai vid failures... Smoke Eating Dancing Twirling 2 or more ppl interacting. Getting unusual camera shots Handing things like guns, swords, forks, instruments. It's really poor for a storyboard artist. You have to always go down the safe boring route.
When it works, it's truly special. Like that woman having difficulty breathing while sleeping, easily a 1080p or 4K shot documentary or movie clip. But many of the others are meh given what we have from the other video models now. I think they waited too long and ended up with more competition than they thought they would
Very high quality but big problems with phisics I think is more important then quality because you can upscale something low but you can't fix something wrong like cars that desapear or drive In the wrong way etc.
@@TheoreticallyMedia yes, as the results are not so good , vs Kling, for example,, so if they put in inside chatgpt, then yea, I will use this .. I build custom GPTs that create already nice prompts,, now will just generate videos inside chatGPT, I will be super happy .. so waiting .. the final shot
There's some good outputs in here for sure-- but yeah-- also, kind of lacking some of the originality and style we've seen out of Minimax and Kling, to be sure. Image to Video in Sora is the thing I'm curious to see.
If Sora releases, its going to be unusable. I mean, look at dal-e -> they've nerfed it into the ground. It used to be the best, but now the output for realistic images sucks. And you can't use a lot of promps, because of censorship -> i wanted to create an image of a king charles spaniel (which is a dog species). Nope, cant do it, because its the name of a person.
I've been saying that for awhile too. DallE, but video. That said, I think you might be able to generate some cool stuff up and use to in a V2V or even Rotoscope stuff out of it. It might have a good spot as an asset generator, for those of us who are trying to push things. For like, Uncle George posting up his "funny" memes on Facebook-- He'll have a lot of fun with Sora.
I tend to think there is a "Sora 2" behind the scenes. I think what we got a glimpse of is like the 6mo old version. That said, we'll never know until they actually ship it!
When Sora was first announced, I was pretty amazed... but then we didn't have access to it... then folks who did released some pretty cool videos and I was stoked again. But after using Kling and Hailuo, I'm not seeing a ton of generations that really blow me away or that I can't get out of Hailuo, prompt and img to vid. I lost all interest in Sora. They're just an old wives tale at this point. The stuff I'm getting with the current platform has already caught up to and in some cases exceed what I've seen in Sora. It's no longer a boogeyman, smh. Way to go OpenAI... Quality, prompt coherence, motion... from what I've been making, Minimax seems to have the upper-hand. I'm only going on what I've seen from Sora, since I can't actually play with it it's had to compare but as far as I'm concerned I don't really need to wait for them anymore. The stuff coming out has gotten better while access to Sora is still a closed party. We don't need their stinkin' party... I'm jammin out to the MiniMAX lol
Haha. Wasn’t me!! Funny with the variety of prompts. You have this moment to play with Sora, and the first prompt you choose is… Ok, lesbians kissing. I mean, that’s a choice! Haha
With no Image to Video, I agree. IF it (eventually) releases with I2V-- it could be a real contender for top throne. As long as it maintains that 1080 output and smooth motion. We'll see...I honestly, don't think they'll release Image to Video.
Looks like Sora pulled back its compute budget to be more realistic, these look worse than the initial hype videos due to possibly less compute used. They're far more grounded in the market of the competitors now.
I think that was the most interesting thing about it. Granted, this is the "turbo" model (supposedly)-- but I think it illustrates exactly where Sora would be for general public use. Granted, there are some extras in here that we still haven't seen. Like, Image to Video...and I will say, that 1080 motion looks really good. But-- overall, yeah. Not much here beyond 10 second outputs, that couldn't be done in what is already available.
Creo imágenes y vídeos con IA ,y solo digo ,sora lo va a romper es y se nota que lo quieren sacar perfecto pero el realismo y fantasía ,las manos es insuperable.
OK, some of it is actually impressive. But in principle I could do most of it with Runway, Kling and Minimax. And, the Sora model has just as much trouble understanding our physical world. Ultimately, as always, AI quality depends on our “prompt quality”. So what? What's all the hype about Sora? I think that OpenAI isn't really interested in Sora as a product and just wants to show that they can do it too, because they can. A few nerds unexpectedly get access to Sora for a few hours and all they come up with are mostly stupid prompts that Sora almost always misinterprets... LOL
I'm still 50/50 on that. Now that we've seen it, a lot of the chatter is that it actually isn't super impressive. To note, I still think it's pretty good-- but, I'm thinking we MIGHT actually get a release sometime next month. But, I also wouldn't be surprised if it never appears.
Sora most videos are not very impressive because I'm using other models that are much better. Also, Sora doesn't have an image-to-video feature and lacks consistent character representation. What's the point of just creating videos? It's just a waste of money.
agreed. To be fair, this is the "turbo" model, meaning there is a bigger/badder model still locked up somewhere. That's the one that likely has I2V and more. But...that's also the one I doubt any of us will be able to play with.
Haha, yeah- it was supposed to be my "day off"-- but I woke up this morning and had a little extra time, so I figured I'd at least pop this up for the historical record. Ok, off to Thanksgiving Weekend! Enjoy your turkey day!!
Just a quick compilation of all the User Generated Sora videos that were recently "leaked" on Hugging Face. I tracked (I think) all of them down, and was able to provide the prompts used. It's kind of a fascinating look at what Sora can and cannot do. No Commentary on this one, just the videos!
Bad news story, IMO
😔
P.S. But I really like the background music in this video!
@@LouisGedo Thanks! This was an old Ambient experiment from back when I was doing more music stuff on the channel. It was a lot of fun to make-- not AI! Fun with Synths!! I can get lost doing this kind of stuff for hours!
There is a new update in google imagen 3, where now you can create images with several aspect ratio options. Maybe it can be an idea for your video content 👍
@@arimuraminami I'll check into Imagen again. I took a look at 3 when it first launched and I wasn't overly impressed. But, that happens a lot at launch. The models always improve.
Were the prompts for these available?
This video is the sora killer.
Haha, best comment so far!
Good one! That and the fact the OPENAI is paying Africans peanuts and giving them PTSD: ua-cam.com/video/qZS50KXjAX0/v-deo.htmlsi=C0ck4SARUPix3LK5 as they detailed in 60 Minutes.
🤣
This thing is clearly not ready yet
My grandfather remembers when Sora was announced.
good times, full of hopes and dreams
🤣
A few of these are amazing, while most look like we’ve taken a step back.
I think what's fascinating about this is when you examine the prompts. I think Sora needs a lot more prompt help than we think it does. We've been getting used to LLMs as an in-betweener with those "magic prompt" enhancers.
Clearly, Sora is in need of that. If only OpenAI had access to a LLM...hmmmm...
Yeah I don't wanna be that guy but I expected way more from openai, especially after like a year, even if this is a turbo model.
Mind that many were just 360-480p.
Is it possible that the lack of coherence is due to the artist not having enough GPU power at their disposal? Maybe these weren't rendered on the OpenAI servers...
I was thinking the same. We've seen hundreds of examples better than the best stuff presented here. Open AI just kicked off the revolution with the presentation of Sora and everyone has been breaking their necks to catch up. And well, they've caught up alright. Real competition is in play. May the best models win.
I've been saying this for months. The reason they won't release sora is because once they do the grift is up. The examples we saw at the beginning of the year were cherry picked and the thing is a slop machine. Things people actually want to watch are never going to be created by text prompts
3 video updates in a row, you are on fire this week 🔥
Haha, this last one was kind of spur of the moment. Onward to Thanksgiving and off the rest of the week! No matter WHAT happens in AI!
(famous last words...)
I applaud Sora for raising the proverbial AI video bar. It’s still closed vaporware so far as I’m concerned, until the public can get its dirty hands on it. Hopefully this changes soon…. Great video 🍻
thanks for putting this together Tim!
The fire throughout these generations is pretty impressive.
You know, Tim, this is a great vibe, I think it will go well to review this video! Sora's work may not seem stunning here, but there is a dynamic 🤗
Agreed-- but, I kinda missed the boat on this one (funny enough, I was getting a new dishwasher that day), and I kind of feel like all the "hot takes" made the rounds.
But, I've got a lot of thoughts on it, and saving them up for the next big Sora news cycle story. Hopefully, that will be the actual release?
Sigh...
I think what people chose to prompt says as much about the state of things right now as the generated videos! :D
Man, I kept thinking the same thing! You have one magic Sora prompt, and…that’s what you went with? Interesting.
I shudder to think what would happen if 3-Wish Genies actually existed. 😂
*
Sora reminds me of that simpons scene where mo steals homers idea and homer is in mos now packed bar screaming "you just lost a customer" .
Mo is shouting back 'i cant here you homer" as the till rings out with the hundreds of punters buying beer.
Haha, YES!! Flamin' Moes!!!
Thanks so much for sharing this. It is the closest we all have to seeing what SORA is actually capable of!
Awesome to hear! Yeah, this was exactly the video I thought we needed. Lots of commentary videos, but I really just wanted to put one up that just showed the footage. No memes, jokes, or blathering. Just the footage.
And, y'know-- obviously we can chat about what we think in the comments!
@@TheoreticallyMedia Straight to the point and straight form the source. No better way than that! 👏👏👏
What's the point? The train is going faster every day and SORA is still at the station, reading the timetable. Among the untouchable models the Meta Movie Gen looks far superior to this and with sound.
With you on that. Funny enough, months before Meta dropped their announcement, the Sora team did an interview with MKBHD where they mentioned sound for Sora 2.
But...typical OAI, they missed the kill shot on the announcement.
rumor has it the dude's still swirling around with the ball, and a basketball basket appears in front of him.
Some of these prompts are hilarious!!!!!
I kept thinking that as I was copy pasting them in. You have one prompt for Sora...and that's what you went with? No judgement! But...really?
This video is great as a sleep aid. I keep dosing off trying to get through it. Snore-a.
Haha, the ambient music I put under it probably doesn't help!
You're great man thanks 🙏
I earned my cape for the day!
Nice to see real videos, not only the top ones they show when try to sell the product.
I hope this makes they be more rumble at pricing.
There are glitchs and fails as in all other models.
Here's the good news: some content was generated without being flagged and blocked! Thanks for the no-comment video and prompts so we can try to absorb it all. I see some limitations, but also more usable content, especially dynamic motion, albeit slow-paced. Wondering about how long it takes to generate content and any cost compared to others charging $95. Hopefully, you can get an official eval!
100%! I figured all the other channels had done their hot takes, but no one actually comp’d everything together. Haha, I did miss my chance for the “hot take” because I had to get a new dishwasher. Haha, glamorous life of UA-cam!
So, I figured this would serve as a good little time capsule for the moment we got to play with Sora.
For sure there are some “eh” outputs, but there’s also a lot of usable stuff in here. And since I was able to track down the prompts, I think it serves as a really fascinating look at what you can and can’t get out of Sora.
😂 last part. Yep noted but appreciated your comp because of all the hot take noise elsewhere. Will look forward to yours none-the-less. Happy Thanksgiving!
Otherworldly, and a little nightmarish. Thanks, Tim.
Happy Thanksgiving to you, too! lol
Thank you! Enjoy your Turkey Day as well! Have an extra piece of dessert as well! It’s been a crazy year, we’ve earned it!
FYI that final prompt is "Aoi Sora", or "Blue Sky", so it did a good job at generating itself!
Ha! Thank you! I meant to run them through a translator, but-- then I would have missed your comment! haha, appreciate it! And that's also, pretty funny!
Wowzen! Bravo you picked a perfect music too… is it by Phillip Glass?
Haha, that's me-- but, kind of aping Phillip! And NOT AI as well. I actually broke out a bunch of synths a year or two ago for a little mini EP experiment. It was a lot of fun!
@@TheoreticallyMedia another Wowzen!! maybe I use one of your tracks on my ai videos :) ... are you on IG ? we can collab 🥰
Thanks for the showcase. Some of these are excellent, some not so much but I also believe it isn't the full model in use. The thing is, even if Sora does release in some form, there's more competition now than when the earlier footage was seen and it's only going to increase.
Question: Now you've seen this footage, do you think that earlier footage was genuine, aside from being heavily cherry picked? I seem to recall a lot of folks claiming it was fake.
Yessssss, can't wait for it to come out 😊
I watched some of that with my mouth wide open, but in a years time we will all be laughing at how terrible it is.🧡
Yup! That’s in part why I wanted to post this without commentary. Now it’s a little time capsule!
How many more "in a years time" will you guys say until you realise this stuff has diminishing returns and doesn't improve linearly?
@@sidbrun_ Who knows? You can't deny that the current evolution of AI is very fast, like it or not. I'm a musician and visual artist, and I say armbrace the new or die.
@@-Deena. Good mindset you have. I don't even have to wish you the best in this AI world because you've already taken the steps to succeed in this new reality. (which we saw coming from a mile away)
@@jojothepro15 Thank you. We all have to remember that at one point oil paint, photography, film, were considered the work of the devil to some, and unless practicing artists embrace new technology, AI will end up producing only clichéd teenage nerd boys sexual fantasies, and no one wants to see that!
Some time ago, I watched a video where you compared Sora as the generative AI video standard to other video creation companies. I mentioned that it felt unfair to include Sora in the comparison because it’s not even available yet. You responded by saying it was like refusing to rate an ARRI Alexa against other cameras simply because we don’t have one in our possession. My point wasn’t just about not having access to Sora-it was that none of us have been able to thoroughly test it like we have with other platforms like Kling, Runway, Hailou, Kaiber, and others. With those, we’ve had the opportunity to explore both their strengths and areas for improvement. In contrast, OpenAI keeps receiving credit for Sora’s potential without even actively participating in the competition. And now they have, and they are lacking.
Id be curious to see a poll, is sira better than the best publicly available ai video out there?
Generator? Currently I think the crowd favorite is Minimax. Unofficial, of course-- but that's just the one that I've noticed people seem to really like the most.
So beautiful 😍 ❤
People are saying that sora isn't better than other models like kling or minimax. But sora is the best on physics
Agreed. The smoothness of motion in characters is fairly stunning. Although, I have to admit, personally I find the T2V outputs to be a bit bland. Image to Video? That's going to be stunning.
AGREED
Have you even watched these clips? There's no evidence that it's the best with physics. Lots of weirdness to be found, if you're paying attention.
Don't forget runway with midjourney sora can't compete with no one yet as it's is not available to the public runway is better as its 80 a month for unlimited minimax and kling don't offer unlimited
If this is the best in physics, we're in trouble
Not excited for sora anymore.
There's way better possibilities to follow.
At this point I don't think I'd use sora if it released tomorrow.
I want Sora's results with runway's render time :)) we still need to figure out how long was the waiting time, kling for example, have gone up from 7 minutes to 13 minutes for PRO users, that's sucks, and highly inefficient, if Sora can bring Runway's render time, they got me!! as for the overall video quality, it looks like the model was mainly trained on National geography channel. the animals are superb, the people,,,, hmm... not so much... but then again, just a few months ago 70% of the people in the comments would argue that it's not real, and open AI are faking "ai generated videos"... so we are in a good place :)
Insane 💯
Imagine that you're in a light sleep, between dreams, just barely able to hear the sounds in the room around you, and someone says something -- describes a scene or event -- and you start dreaming about it. That's a prompt. And that's often about as good as Sora's prompt adherence is. What we need is something that reliably follows prompts, in their entirety, and understands the meaning of the prompt to the degree necessary to accurately follow it. Sora isn't there yet. As such it's limited in its utility for things like movie making. It's hit or miss. Given how expensive it is, how much energy is consumed, how much cooling is required, the compute resources needed, we really can't afford a technology where it produces a bunch of unuseable results for every one that does actually adhere to the prompt with sufficient accuracy. They need to fix this before Hollywood is doomed. I trust they will succeed, or someone else will, and only then will Hollywood be doomed -- or at least will have to adapt its processes and techniques.
5:30 is amazing for anime short.
10:42 That was really funny 🤣
12:14 I thought he would throw the ball
13:38 The prompt was literally "blue sky", and it really was blue sky. dude missed the opportunity to make a cool prompt 🤷♂
Those guys couldn't have given openAI better marketing assistance if they tried. let's see if sam jumps on it and uses it
Fully agree. If this ISN'T a PR stunt, because I'm so turned around, there's a 35% part of me that thinks it might be-- then yeah, this was a gold mine for them.
Just curious: Did you upscale these? The twitter versions did not look near as good…😮 And I think the Sora turbo model understands anime a lot better than the competition. Imagine what the prime will be capable of? Also, loved the music!
I got the source files. There's still YT compression on them, but Twitter's compression is REALLY bad.
oh, and thanks about the music!! Yeah, that was an old experiment I did awhile back (not AI generated)-- glad it finally found a home!
@@TheoreticallyMedia Yeah, I know what you mean. All this AI and the energy it takes to keep up with it has taken its toll on me too. I am a US Vet (in the same age range) who back in the day used to record all my music by myself and mix/master, do all the release legwork and artwork, etc. (no AI of course!) and now somehow I can't find the time to pick up my old Les Paul anymore...Lmao! 🥲
Honestly, I don't think SORA is better than Runway, Kling, Hailuo, or the like. It would be great if Open AI can make SORA more affordable than others.
Well not exactly what i was expecting when you said rumours were about suggesting Sora would be released soon.....hehe
If you pause and read the prompts, it is not exactly showing what was asked in most cases.
Oh great, now I'm hungry for an i-Bupler.
Tim Cook in the kitchen! Preorder now for $2999. Well, that's the "Pro" bupler. You can get an Bupler Air for much cheaper.
@@TheoreticallyMedia yeah but the air model is so much thinner
А когда будет релиз?
Give review of New Luma Machine version
Coming soon!
most people doing pic to vid now days much more control but still the jank if you look hard enough so I wonder how well it handles with pics/vids hopefully it has video to video
So.. This is why they restricted it for now..
When it comes out it will be #1, and Hailuo AI #2
Just amazing 🎉
Prompt adherence not great
I think it needs a LOT more coaxing from the prompt. Simple prompts (for now) aren't cutting it w/ Sora.
Hmmm, if only OAI had a LLM to help. Shame they don't...haha.
Nice music!!! Made watching these really feel atmospheric. Have to say tho... Im not terribly impressed with the Sora generations. A couple were great... particularly some of the anime... but really Im seeing a lot better from other generators. and yes i do know its Sora Turbo... but still.
와 역시 Sora가 퀄리티가 좋네요. 하지만 한편으로 다른 AI도구들도 많이 따라온것 같습니다 :)
I told everyone that if they thought a company that was miles ahead has just been twiddling their thumbs for a year then they would all be proven wrong. And well, yup. Especially if this is just turbo. A lot of these look a year old as well - the proper Sora 3 is gonna be amazing - this also shows that SORA is the most flexible model in terms of different topics it can do where as kling minimax etc excel at certain things and then completely flop on others.
In Feb when we first saw Tokyo Woman, if you told me we (still) wouldn't have Sora by then...I'd sadly nod. But if you told me we'd have 20 other versions that were just as good-- and in some cases better? I'd be STOKED.
It's not even December yet, and I'm really happy with where we are.
sora is scam and urban legend. It will never be released
@@jzwadlo
Ai vid failures...
Smoke
Eating
Dancing
Twirling
2 or more ppl interacting.
Getting unusual camera shots
Handing things like guns, swords, forks, instruments.
It's really poor for a storyboard artist.
You have to always go down the safe boring route.
OpenAi has been stuck for months because its staffed with wokesters all stabbing each other in the back and then jumping ship
When it works, it's truly special. Like that woman having difficulty breathing while sleeping, easily a 1080p or 4K shot documentary or movie clip. But many of the others are meh given what we have from the other video models now. I think they waited too long and ended up with more competition than they thought they would
The prompt was for a “slightly fat lady” I think it went overboard.
Of course this doesn't demonstrate many of the much anticipated sora features like consistent characters and video to video
Not soon enough ….
Isnt sora for music ai?
That’s Suno! Ha, never realized how close names were!
Wow the end is near!
Im telling u!!
Why did nobody try copyrighted characters 🤬
Thanks man :)
I earned my cape for the day!
Very high quality but big problems with phisics I think is more important then quality because you can upscale something low but you can't fix something wrong like cars that desapear or drive In the wrong way etc.
Full movies in 5-10 years
9:44 reminded me of J.Jonah Jameson (J.K Simmons) from Spiderman 2002 when he was laughing like this
will this work inside the chatGPT?
maybe eventually? If they ever release it.
@@TheoreticallyMedia yes, as the results are not so good , vs Kling, for example,, so if they put in inside chatgpt, then yea, I will use this .. I build custom GPTs that create already nice prompts,, now will just generate videos inside chatGPT, I will be super happy .. so waiting .. the final shot
I have the feeling that Sora won‘t be able to compete against other (already released) models.
There's some good outputs in here for sure-- but yeah-- also, kind of lacking some of the originality and style we've seen out of Minimax and Kling, to be sure.
Image to Video in Sora is the thing I'm curious to see.
But its not done yet its not supposed to be leaked
If Sora releases, its going to be unusable. I mean, look at dal-e -> they've nerfed it into the ground. It used to be the best, but now the output for realistic images sucks. And you can't use a lot of promps, because of censorship -> i wanted to create an image of a king charles spaniel (which is a dog species).
Nope, cant do it, because its the name of a person.
I've been saying that for awhile too. DallE, but video. That said, I think you might be able to generate some cool stuff up and use to in a V2V or even Rotoscope stuff out of it.
It might have a good spot as an asset generator, for those of us who are trying to push things.
For like, Uncle George posting up his "funny" memes on Facebook-- He'll have a lot of fun with Sora.
Why aren't these tools available to general public, but only non-consistent and expensive garbage?
Some videos are pretty good :)
Honestly, current platforms like KlingAI are much better than Sora. They need to enhance it before releasing it.
I tend to think there is a "Sora 2" behind the scenes. I think what we got a glimpse of is like the 6mo old version. That said, we'll never know until they actually ship it!
When Sora was first announced, I was pretty amazed... but then we didn't have access to it... then folks who did released some pretty cool videos and I was stoked again. But after using Kling and Hailuo, I'm not seeing a ton of generations that really blow me away or that I can't get out of Hailuo, prompt and img to vid. I lost all interest in Sora. They're just an old wives tale at this point. The stuff I'm getting with the current platform has already caught up to and in some cases exceed what I've seen in Sora. It's no longer a boogeyman, smh. Way to go OpenAI... Quality, prompt coherence, motion... from what I've been making, Minimax seems to have the upper-hand. I'm only going on what I've seen from Sora, since I can't actually play with it it's had to compare but as far as I'm concerned I don't really need to wait for them anymore. The stuff coming out has gotten better while access to Sora is still a closed party. We don't need their stinkin' party... I'm jammin out to the MiniMAX lol
"Two japanese lesbians kissing"
Woah woah woah slow down there theo
Haha. Wasn’t me!! Funny with the variety of prompts. You have this moment to play with Sora, and the first prompt you choose is…
Ok, lesbians kissing. I mean, that’s a choice! Haha
@@TheoreticallyMediaahahahaha i was shocked to see sora generate such thing
SORA today is just another AI among many others for video generation, sometimes even inferior to what we have on the market.
With no Image to Video, I agree. IF it (eventually) releases with I2V-- it could be a real contender for top throne. As long as it maintains that 1080 output and smooth motion.
We'll see...I honestly, don't think they'll release Image to Video.
I could really go for a bupler right now.
I'm holding out for the Bupiler Air. It's $999, sure-- but it's SO THIN!
Double burplur with cheese
Looks like Sora pulled back its compute budget to be more realistic, these look worse than the initial hype videos due to possibly less compute used. They're far more grounded in the market of the competitors now.
very disappointed in this, it's barely better than current competitors.
I think that was the most interesting thing about it. Granted, this is the "turbo" model (supposedly)-- but I think it illustrates exactly where Sora would be for general public use.
Granted, there are some extras in here that we still haven't seen. Like, Image to Video...and I will say, that 1080 motion looks really good.
But-- overall, yeah. Not much here beyond 10 second outputs, that couldn't be done in what is already available.
'why are we doing this' lol
My favorite prompt of all of them!
Whats the deal with the big holdup in releasing this to the public like Kling, Runway, Luma, etc? Is this just a marketing ploy to create hype?
Creo imágenes y vídeos con IA ,y solo digo ,sora lo va a romper es y se nota que lo quieren sacar perfecto pero el realismo y fantasía ,las manos es insuperable.
Sora who ?
Knock Knock! Who's There?
Sora! Sora Who?
Sora-Prise, it's a leak!
@ no need to knock the doors are Open ai man
Very soon reality won't be able to compete.
Leakage is a sign of adolescence. Maybe AI-dolescence.
OK, some of it is actually impressive. But in principle I could do most of it with Runway, Kling and Minimax.
And, the Sora model has just as much trouble understanding our physical world. Ultimately, as always, AI quality depends on our “prompt quality”.
So what? What's all the hype about Sora? I think that OpenAI isn't really interested in Sora as a product and just wants to show that they can do it too, because they can.
A few nerds unexpectedly get access to Sora for a few hours and all they come up with are mostly stupid prompts that Sora almost always misinterprets... LOL
It will never released
I'm still 50/50 on that. Now that we've seen it, a lot of the chatter is that it actually isn't super impressive. To note, I still think it's pretty good-- but, I'm thinking we MIGHT actually get a release sometime next month.
But, I also wouldn't be surprised if it never appears.
@TheoreticallyMedia why do u think so?
Strange, the sora videos from half a year ago looked much better. Beside of maybe some of the animal clips kling seems to be one step ahead.
The woman in tokyo more detailed and the scene this time i wonder if this sora 1.5 or turbo
None of them are better than Kling, RunwayGen3, or MiniMax.
prompt OK image not generate
Wake me up when we can use it😴
Sora most videos are not very impressive because I'm using other models that are much better. Also, Sora doesn't have an image-to-video feature and lacks consistent character representation. What's the point of just creating videos? It's just a waste of money.
If it was that good, why not let the public play with it. Invideo looks better
TOO FECKING LATE SORA!!! I wouldn’t be surprised if this an elaborate Theranos investor type scam
Never too late
I dunno.. maybe I waited too long. Paris syndrome
It turns out to be what everyone has been waiting for, but it's not much different from Kling AI 🤦♂️
agreed. To be fair, this is the "turbo" model, meaning there is a bigger/badder model still locked up somewhere. That's the one that likely has I2V and more.
But...that's also the one I doubt any of us will be able to play with.
c'est joli mais on retrouve les memes defauts qu'ailleurs. pour l'instant le combo runway kling minimax fait l'affaire. Rien de nouveau sous le soleil
Coke getting in early
Haha- Right? Two Coke videos in this supercut. They just got some free advertising!
Awesome
This might just be the lightweight model. Hoping the high-quality model will be much better than this 😊
Two Japanese lesbians kissing prompt took me out lol 😂😂😂
We miss the tim commenting! good video but never do this again haha
Haha, yeah- it was supposed to be my "day off"-- but I woke up this morning and had a little extra time, so I figured I'd at least pop this up for the historical record.
Ok, off to Thanksgiving Weekend! Enjoy your turkey day!!
Looks in line with Kling...the slow motion just ain't it
8:15 real life Tom and Jerry
usual slow motion taking up double time of what should be a much shorter clip. bla bla.... kling already beats this