Bytedance's model's most impressive feature that NOBODY ELSE has is the ability to cut while remaining consistent. Mind blowing and I'm surprised you didn't mention it. It's by far the standout feature of this model
Same here, the Adobe Photoshop that I actually paid for-- eventually revoked the license for it which from what I understood was supposed to be good for a lifetime, and now I refuse to use their software.
Honestly I am really looking forward to whatever the boys at Black Forest Labs are cooking. Especially once we can use it in conjunction with Flux 2.0 and it's fine tunes.
Hey Matt, any plans to do a video on the new Gemini live voice assistant? I heard it's just been released, and it's free for everyone to use now. I looked it up online, and it seems legit -could be worth checking out!
Sora, once a promising AI video generator, is clearly getting left behind as newer models have rapidly overtaken it. Platforms like Runway Gen 3, Helu AI, and Luma Labs are pushing the boundaries with impressive features like faster generation times, higher-quality visuals, and advanced motion tools. Sora’s lack of access and underwhelming quality in comparison to these new players highlights how quickly it has fallen behind in the AI video generation race.
Im pretty sure theyre still miles ahead of everything out now pal lol. Channels that claim they have been 'caught up to' have literally 0 idea what is going on and you'll be able to quote me on that. Literally all these models a year on are BARELY catching up to what Open AI CHOSE to show the public. If you think that's where Sora was in development at that point, it wasn't. And if you think they have just been twiddling their thumbs since then, wrong again. They don't need to release anything till when they need to as they make plenty money from Chat GPT man. Whenever Open AI is ready, they will nuke everything out there I guarentee.
Too early to say. OpenAI also came out with a logical thinking model although not perfect, they are ahead of the competition overall in AI again. Sora bit of let down for not coming out but again took strawberry 🍓 about a year to come out after training too. Sora might come out with a much more advanced and sophisticated version when it does come out. It’s also a world simulator and supposedly you can move around in the worlds it generates so it’s different from most of the stuff out now from what I heard.
They had the model before anyone else. They could have destroyed everything but this stupid obsession with "safety" (A.K.A. making sure not to offend anyone) screwed them. It's the same with 4o's advanced voice mode being nerfed to a shadow of what they showed us months ago in those demos.
@reezlaw Agreed. I suspect that when China embraces the chatbot market, they will kick openai with less of the "offend me not" culture. Google live is even worse.
All of the examples are a single prompt. That’s sort of why all the scenes aren’t perfect yet on Bytedance Seaweed, is what it’s called. The scary scene, and the robots with lasers are a single prompt, with the cuts described in the prompt.
Good evening, AI enthusiasts! Yes, I do agree that video generators are evolving a LOT, especially with Pika Labs recently! Sad that we couldn’t see most of the generations you tried to make with it, though.
Obsolete and given up on sharing anything online because nobody will believe you're a real person, and half or more of the feedback you get will just be AI/bots anyway.
13:37 imagine, 2042, AR/XR Glasses are in their 6th generation, compute power is overflowing everywhere, prediction + video gen models gives realistic depictions of the near future before it happens right within your glasses.
The interesting thing about the Bytedance video generation is that the 2 cuts in the video are all in a single prompt. It’s not 3 clips brought together. Same in the rest of the examples. You will be able to have multiple cuts in the same single generation, up to 10 secs in a prompt. Supposedly it’s going to be able to do a 2 minute single prompt.
chinese ones are better because they train on recent movies and popular media rather than just stock video and and youtube content.. they don't care about copyright
crazy times! when i first saw the sora video i figured wed get that level in like end of the year but now we got many options and all since then too crazy!
I have seen it reported that OpenAI is retraining Sora, because it took to long to generate videos. The retraining is probably related to the upcoming "blueberry" image models, which beat Flux Pro and also generate faster than Flux as well. When Sora does release, it will basically be Sora v2 by that point.
will have to try some of the others, kling is great for its quality, and ltx is neat for its multiple scenes or same character generator. can literally make a whole story with the same person. if they just upped the quality of it a little bit some distortion still sometimes.
I'm waiting for these video generators to start working side by side with software like BLENDER. I'd love to plug a few Blender creations into these video generators to create something I really want. That's when things will get really amazing.
Honestly, I watched that full video and there’s so much information. I can’t even really ascertain what your favorite was in the end. Always good to do a quick summary at the end.
Is there an image generator that can stick to the same subject across multiple different generations? I'm asking for a children's book to have consistent characters. Thanks.
Cool stuff for sure, still looking for a way to make longer form consistent 2d animations. I really wanna produce a mini series and have the voice actors and script but I want to create the animations with AI. Nothing yet on the market it seems.
Sora is awesome. I used it for my spaghetti short. It has its own look. For the project im doing im playing a character (Mvikeli v3.5.7) who is an AI and i find Sora gets it.
It's amazing how things are impressive it's going to be really amazing about how the original artists are getting their artworks stolen poetry and so forth and all the lawsuits will be created from it plus the people that are making AI have in their stuff stolen from other ai and having it sold and not even getting copyright credit for it or residual income
I tried Pika 1.5 this morning but it still hasn't made the video. I hope it works soon because I quit Luma. I had a DOZEN bad renders in a row and when I wrote to them they never responded. I currently use Pika for lip sync but I have Kling and might try their lip sync soon
Hope you can compare Kling 1.5 to this new Bytedance model once its out. And is it maximum 10 seconds per clip like Kling, or scaling up to maybe 20 sec?
We've come far enough out of the uncanny valley that you notice inconsistencies in the surroundings as much as you do the people. I didn't think we'd be there until 2030, but damn. I know the people aren't perfect, the wrinkles in the face seem to have their own muscles vs being fat and skin being pulled by muscles, if that makes sense.
Dude I started a podcast using notebookLM to help me digest information a little better since I have trouble focusing. Let me tell ya, I'm SHOOK! I made a podcast called god is in the code and I was like NOTHING IS REAL.
Super useful video - I don't have time to try all these new models so it's great to see your work. I'm currently trying the new Kling-AI lip-sync features by importing dialog generated in Speechify but my bit-brain is still reeling from learning the Kling-AI motion-brush features - phew.
Cool😊 so minimax will soon have image to video? This week? Bytedance company one is that the Seaweed AI with consistent characters and rotate camera 180 degrees and character has still consistent look and clothes? Guys please vote/comment what's the best video ai. Thanks.🎉😅
Which one works with a normal effing signup and actually works lol I’m watching but damn I tried em all past few weeks mostly junk. Chinese apps, need to pay to even try and like runway, no where near ever what I request
"Something you'd see on Netflix for sure" hmm... 2026 they'll have bought people's voices and likeness and just generate shows vs paying a whole team. Couple writers, the whole budget after is just paying people to craft the scenes in AI. Maybe it's already happened when they can't get the expression "just right", just ai their face to fit the scene lol. Makes less reshoots.
Cool pretty soon channels like yours will be drowned out by thousands of AI ripoff channels made by taking the data and content of channels like yours. SO excited for the days when there's no point in trying to express ourselves at all because everybody will think we are just AI, and half or more of the interactions we get will be AI/bots. It's gonna be SO GREAT!
19:30 "realistic physics of billiard balls"... Have you ever seen how billiard balls move? I mean, how do these move if the numbers stay towards the camera and the only rotation is around the axis perpendicular to the billiard's surface 🤣
Hailuo forces me to give my phone number and though im not overcautious i already have enough calls from India, Nigeria, etc recently since i used my number for another AI site. Kling - as so much others - wont generate for me. at the beginning few weeks ago before 1.5 came out i was lucky to get 3 short generations. From then on - no chance, over a dozen tries für days on 99% then "process failed". I would even pay if they gave me at least ONE try for the 1.5 model - but they locked that behind paywall anyway. Adobe... well i think we all know about adobe pricing. Im a bit tired of hundreds of handpicked videogenerations from all companies that wont work in reality. So i will probably try one paid month with kling, though a lot of ppl say that even paid they cant generate videos with kling. For AI logins that force telephone number like Hailu i will try to get a separate sim/number im afraid. We will see, but other than music- and image-generation i think video is still not even close to be used professionally.
Sora is rapidly becoming 'back in the day'
And it's not even out yet! (And probably never will be, frankly).
The day that never was
@@vainezaiven6677When OpenAI video generation comes out it likely will be the best on the market. They're nearly always one step ahead of the rest.
You're so impatient.
@@jamesjonnes At this point I highly doubt they'll release it to the public.
Bytedance's model's most impressive feature that NOBODY ELSE has is the ability to cut while remaining consistent. Mind blowing and I'm surprised you didn't mention it. It's by far the standout feature of this model
Adobe is dead to me. I automatically zone out as soon as I see/hear/smell anything Adobe.
I skipped that part
Pretty much
Same here, the Adobe Photoshop that I actually paid for-- eventually revoked the license for it which from what I understood was supposed to be good for a lifetime, and now I refuse to use their software.
Why?
Agreed I think our entire tech industry is better without this scab of a company and I hope AI influencers start omitting it
"Assbolutely deserves to be in a museum" .... I see what ya did
Don't inflate dogs, guys.
As long as we can inflate frogs
Could you please instead do a systematic review test of the ones that are available? Same prompts and compare the results. Thx!
Matt: This video generation technology is improving rapidly!
Also Matt: I can't wait till this technology just improves rapidly.
I hear ya!
Never enough
12:18 a fine beautiful work of art
kling had 2 videos of mine at 99% for a week
I gave up on kling
@@ShiroAisan i feel you
@@wrillywonka1320 no reason to kling-on any longer.
had to make a star trek joke, sorry.
@@jetjazz05 🤣🤣😂😂 that was solid lmao
Honestly I am really looking forward to whatever the boys at Black Forest Labs are cooking.
Especially once we can use it in conjunction with Flux 2.0 and it's fine tunes.
Hey Matt, any plans to do a video on the new Gemini live voice assistant? I heard it's just been released, and it's free for everyone to use now. I looked it up online, and it seems legit -could be worth checking out!
Sora, once a promising AI video generator, is clearly getting left behind as newer models have rapidly overtaken it. Platforms like Runway Gen 3, Helu AI, and Luma Labs are pushing the boundaries with impressive features like faster generation times, higher-quality visuals, and advanced motion tools. Sora’s lack of access and underwhelming quality in comparison to these new players highlights how quickly it has fallen behind in the AI video generation race.
They didn't release Dalle 1 to the public either. They are probably waiting til sora 2 is ready.
@@jaredf6205 but why anounce sora if your going to wait until sora 2
Im pretty sure theyre still miles ahead of everything out now pal lol. Channels that claim they have been 'caught up to' have literally 0 idea what is going on and you'll be able to quote me on that. Literally all these models a year on are BARELY catching up to what Open AI CHOSE to show the public. If you think that's where Sora was in development at that point, it wasn't. And if you think they have just been twiddling their thumbs since then, wrong again. They don't need to release anything till when they need to as they make plenty money from Chat GPT man. Whenever Open AI is ready, they will nuke everything out there I guarentee.
@@levonkenney No idea, I don't know why they did it with Dalle either but they are known for weird decisions.
kling still the best though
It's awesome to see how fast AI video generation is progressing. Thanks for the useful market overview and update on new models.
With all these video AI solutions, is Sora even a thing? I think openai may have bitten off more than they can chew at this point.
Too early to say. OpenAI also came out with a logical thinking model although not perfect, they are ahead of the competition overall in AI again. Sora bit of let down for not coming out but again took strawberry 🍓 about a year to come out after training too. Sora might come out with a much more advanced and sophisticated version when it does come out. It’s also a world simulator and supposedly you can move around in the worlds it generates so it’s different from most of the stuff out now from what I heard.
At this point I think it's highly unlikely Sora is ever released to the general public.
They had the model before anyone else. They could have destroyed everything but this stupid obsession with "safety" (A.K.A. making sure not to offend anyone) screwed them. It's the same with 4o's advanced voice mode being nerfed to a shadow of what they showed us months ago in those demos.
@reezlaw Agreed. I suspect that when China embraces the chatbot market, they will kick openai with less of the "offend me not" culture. Google live is even worse.
@@vainezaiven6677 They released Dall-E 2 with equal hype and excitement, we should also get Sora somehow at this point
PikaLabs... light years behind the competition.
hahaha
Do you know what a light year is?
Still waiting on Midjourny video
All of the examples are a single prompt. That’s sort of why all the scenes aren’t perfect yet on Bytedance Seaweed, is what it’s called. The scary scene, and the robots with lasers are a single prompt, with the cuts described in the prompt.
12:59 "They know that's what people are going to be generating most." Butts? Yeah... probably accurate.
Good evening, AI enthusiasts! Yes, I do agree that video generators are evolving a LOT, especially with Pika Labs recently! Sad that we couldn’t see most of the generations you tried to make with it, though.
Pika is so back
17:33 For the record, "666" has nothing to do with occultism in China.
Can't wait to find out where we would be in one year from now..
Obsolete and given up on sharing anything online because nobody will believe you're a real person, and half or more of the feedback you get will just be AI/bots anyway.
Thanks for the update Matt, some impressive stuff there. Look forward to seeing those remaining ones as and when
why promote sora if all is hype and unavailable...Free and open source is for the PEOPLE
They’re inflating the dogs. They’re squishing the cats.
Incredible progress of video models. I think very soon new movies will be created not months and years but just days.
13:37 imagine, 2042, AR/XR Glasses are in their 6th generation, compute power is overflowing everywhere, prediction + video gen models gives realistic depictions of the near future before it happens right within your glasses.
The interesting thing about the Bytedance video generation is that the 2 cuts in the video are all in a single prompt. It’s not 3 clips brought together. Same in the rest of the examples. You will be able to have multiple cuts in the same single generation, up to 10 secs in a prompt.
Supposedly it’s going to be able to do a 2 minute single prompt.
But how to remain consistent across multiple prompts? Feed it images from the previous generation?
12:14 GYAT DAMN, WE ARE TRULY LIVING IN THE FUTURE!
LMFAO
GYAT YOU CHOSE CORRECTLY
chinese ones are better because they train on recent movies and popular media rather than just stock video and and youtube content.. they don't care about copyright
crazy times! when i first saw the sora video i figured wed get that level in like end of the year but now we got many options and all since then too crazy!
AI video generation on your own machine exists now with ComfyUI.
Yes and It's horribly bad compared to those models.
How? Do you have more info?
I have seen it reported that OpenAI is retraining Sora, because it took to long to generate videos. The retraining is probably related to the upcoming "blueberry" image models, which beat Flux Pro and also generate faster than Flux as well.
When Sora does release, it will basically be Sora v2 by that point.
Everyone, including the experts, keep underestimating how fast AI is improving and evolving.
How are you verified with under 100K subs???
will have to try some of the others, kling is great for its quality, and ltx is neat for its multiple scenes or same character generator. can literally make a whole story with the same person. if they just upped the quality of it a little bit some distortion still sometimes.
No way vidu is a fan favorite but runway is middle of the pack.
I'm waiting for these video generators to start working side by side with software like BLENDER. I'd love to plug a few Blender creations into these video generators to create something I really want. That's when things will get really amazing.
Did you catch the news from Liquid AI as well? Aparently will be really good both for LLMs and video.
Honestly, I watched that full video and there’s so much information. I can’t even really ascertain what your favorite was in the end. Always good to do a quick summary at the end.
which model is good for "lip to sync"?
sora 2 will probably be an interactive text to audio/visual LOD chain of imagination world model game/movie engine with scarjo narration
Is there an image generator that can stick to the same subject across multiple different generations? I'm asking for a children's book to have consistent characters. Thanks.
I keep forgetting how exciting this is
Hey, Matt, can you timestamp any Adobe sections so we can skip them immediately? 🙂
Yep, you got the 180 right. That's frame rate = half of shutter speed :)
Everyone should be able to create their own movies and cartoons with their thoughts alone
Cool stuff for sure, still looking for a way to make longer form consistent 2d animations. I really wanna produce a mini series and have the voice actors and script but I want to create the animations with AI. Nothing yet on the market it seems.
@MattVidPro AI we know you were talking about that lady in the museum. Can’t fool us. 😂
Tim's my dude. Thanks for another great video Matt.
I haven't seen an AI video generator that produces anything good at all.
Wait one year
attempting to edit a video currently and keep up with this at the same time lmao the struggle is real
as always bringing the best stuff !! thank you mate
Sora is awesome. I used it for my spaghetti short. It has its own look. For the project im doing im playing a character (Mvikeli v3.5.7) who is an AI and i find Sora gets it.
It's amazing how things are impressive it's going to be really amazing about how the original artists are getting their artworks stolen poetry and so forth and all the lawsuits will be created from it plus the people that are making AI have in their stuff stolen from other ai and having it sold and not even getting copyright credit for it or residual income
30 minutes videos??? Why?? I’m sorry but I can’t spend half an hour just watching your video.
Do what everyone does. 2x speed, is now 15 mins, done.
ADHD
cog is awesome, open source, can be used really well
patience my younger friend: thanks for video
I tried Pika 1.5 this morning but it still hasn't made the video. I hope it works soon because I quit Luma. I had a DOZEN bad renders in a row and when I wrote to them they never responded. I currently use Pika for lip sync but I have Kling and might try their lip sync soon
Hope you can compare Kling 1.5 to this new Bytedance model once its out. And is it maximum 10 seconds per clip like Kling, or scaling up to maybe 20 sec?
Bytedance seems to be on par with Sora, very exciting stuff and I heard it will release after elections so exciting times ahead!
We've come far enough out of the uncanny valley that you notice inconsistencies in the surroundings as much as you do the people. I didn't think we'd be there until 2030, but damn.
I know the people aren't perfect, the wrinkles in the face seem to have their own muscles vs being fat and skin being pulled by muscles, if that makes sense.
Dude I started a podcast using notebookLM to help me digest information a little better since I have trouble focusing. Let me tell ya, I'm SHOOK! I made a podcast called god is in the code and I was like NOTHING IS REAL.
Bytedance's model is incredible, those -ass- art physics though
Super useful video - I don't have time to try all these new models so it's great to see your work. I'm currently trying the new Kling-AI lip-sync features by importing dialog generated in Speechify but my bit-brain is still reeling from learning the Kling-AI motion-brush features - phew.
Cool😊 so minimax will soon have image to video? This week? Bytedance company one is that the Seaweed AI with consistent characters and rotate camera 180 degrees and character has still consistent look and clothes? Guys please vote/comment what's the best video ai. Thanks.🎉😅
Any way to try those new
PixelDance model and Seaweed model ?
Which one works with a normal effing signup and actually works lol I’m watching but damn I tried em all past few weeks mostly junk. Chinese apps, need to pay to even try and like runway, no where near ever what I request
Dnag pika ok it was time for them to update cant wait try it out
Sora took too long to get to us and people filled the gaps. Its basically equal to a pc in the 80s
What happened to Stability's video model?
LOL! OMG! ONE OF YOUR FUNNIEST VIDEOS, MAN!
New in Pika: inflate your dog! Or melt our servers! :D
Is there one with an option to create a loop of 20 seconds or so?
I will not get my mind out of the gutter (:
Thank you for the exhaustive overview. 🙏
NEED FOSS LIST!!!
Chat GPT advanced voice model no available for everyone, free users too ( but 10min/month) :(
this nice vid man , i like compilation of techs , eeeyaa
Love the Axolotl generation.
“The tear drops look like hot glue” 😂
melting looks like After Effects preset
Glue? It looks like her eyes are peeing
Please let the whole videos play, 😅
The melting was pretty weak. Did not look anything like the demo.
Thank you.
Pika actually lookin' good.
great stuff as always matt
Who's ur fav AI "art" critic...?
"Something you'd see on Netflix for sure" hmm... 2026 they'll have bought people's voices and likeness and just generate shows vs paying a whole team. Couple writers, the whole budget after is just paying people to craft the scenes in AI. Maybe it's already happened when they can't get the expression "just right", just ai their face to fit the scene lol. Makes less reshoots.
I have access to sora I spent a good amount of money on their packages 😂🎉 can anybody tell me how to extend 10 seconds?
12:15 - 12:30 hahahah yes.
Parker Crossing
Bro new video ganrator coming this weeek 😂
Its called curling by the way 😅
that's curling, not shuffleboard :)
25:51 Don't show Pyro this.
get it to him ASAP
Cool pretty soon channels like yours will be drowned out by thousands of AI ripoff channels made by taking the data and content of channels like yours.
SO excited for the days when there's no point in trying to express ourselves at all because everybody will think we are just AI, and half or more of the interactions we get will be AI/bots. It's gonna be SO GREAT!
Those racoon eyes, Matt you need to take some rest ❤4rmZambia 🇿🇲
Use video chapters.
19:30 "realistic physics of billiard balls"... Have you ever seen how billiard balls move? I mean, how do these move if the numbers stay towards the camera and the only rotation is around the axis perpendicular to the billiard's surface 🤣
Hailuo forces me to give my phone number and though im not overcautious i already have enough calls from India, Nigeria, etc recently since i used my number for another AI site.
Kling - as so much others - wont generate for me. at the beginning few weeks ago before 1.5 came out i was lucky to get 3 short generations. From then on - no chance, over a dozen tries für days on 99% then "process failed". I would even pay if they gave me at least ONE try for the 1.5 model - but they locked that behind paywall anyway.
Adobe... well i think we all know about adobe pricing.
Im a bit tired of hundreds of handpicked videogenerations from all companies that wont work in reality.
So i will probably try one paid month with kling, though a lot of ppl say that even paid they cant generate videos with kling.
For AI logins that force telephone number like Hailu i will try to get a separate sim/number im afraid.
We will see, but other than music- and image-generation i think video is still not even close to be used professionally.