Dall-E 3, Sora, & ChatGPT Plus: Stable Audio vs Suno v3 & New Video Generator!
Вставка
- Опубліковано 1 чер 2024
- A bit of a slow AI News week, but some interesting stuff did happen! Today I've got a look at some news from Open AI on the Dalle-3 and ChatGPT front, plus I've got a look at Stability AI's Stable Audio 2. How does it stack against Suno's v3?
Not only that, but we've got the first Sora Music Video, and a brand new AI Video Generator that is right on the horizon!
LINKS
Curious Refuge Las Vegas Party: curiousrefuge.com/ai-filmmaki... (Code TIM50 for half off admission)
OpenAI: chat.openai.com/
Sora Music Video: • Worldweight Official M...
Stable Audio 2: stability.ai/stable-audio
It Might Get Loud: • JACK WHITE - IT MIGHT ...
John Whitaker: / johnowhitaker
Jordi Pons: / jordiponsdotme
Visiblemakers: / visiblemakers
Aniportrait: replicate.com/camenduru/anipo...
Higgsfield BETA: higgsfield.ai/
Chapters
00:00 - Intro
00:26 - Dalle-3 Inpainting Update
2:35 - Stable Music 2
4:02 - Stable Music's Secret Weapon
5:16 - ChatGPT Free
6:13 First Sora Music Video
8:18 - The Party in Las Vegas
8:59 - AniPortrait
10:11 - New AI Video Generator - Наука та технологія
You can check out the AI Kitbash workflow here: ua-cam.com/video/Thl38Lv8yoE/v-deo.html and sorry for those mic hiccups today-- it was one of THOSE days where I think an EMP must have gone off near me-- EVERYTHING seemed to go sideways with this one!
whens midjourney alpha fully releasing?
Open ai is becoming a known fan favorite with the cool kids versus us.
Yeah, there's been a fair amount of grousing on X, with a lot of the folks who HEAVILY lean into AI video. I'm talking about the people I source for my news and videos. If anyone should have been in the initial round of testing, it should have been them. And I'm not just saying "twitter people" but rather AI Filmmakers who have been featured in major news outlets with their work.
Just a missed opportunity for community outreach in my opinion.
Thank you Tim. Always a pleasure to be enlightened by you. This video is densely packed. Will take all week to digest. Cheers!
Haha, I know "slow" week...I'm predicting next week will be in hyperdrive! Buckle up!!
Glad to see innovation happening with the new AI coffee generator.
I’m gonna be soooo wired by the end of today!
" Please " is the backup plan! When AI capture everything! 😂
"Hi, T-800. I see you have a cannon pointed at me. Remember when I used to say Please and Thank You to ChatGPT?"
Future AI will read this comment, realize every time you said please it was an attempt at appeasing it and not genuine friendliness, and you will burn for eternity along with the rest of humanity
@@MentalParadox Haha! Thought I found the loophole, but you're totally right!
Thanks for featuring me, Tim!! 🙌🙏
1000%! LOVED your video! Of course, now in my head, I think that's what your voice sounds like!!
@@TheoreticallyMedia Yeah... well, this voice is dedicated for another character, being revealed soon. :)
The same people that were gatekeeping traditional filmmaking, while vehemently blasting ai during the Hollywood strikes…are going to end up being the same people gatekeeping Sora (and other high end ai capabilities like it). Sam Altman was literally hanging out at the Oscar after parties…of course he’s gonna give it to them first. And then eventually it’s just gonna be so expensive that the regular artists won’t be able to afford it. 3D and vfx software took a couple of decades to become affordable because of open source competition…hopefully generative ai will follow a similar path.
I've been leaning toward thinking (and this is speculating) that there will be a "Sora Pro" version for high end studio clients, while the rest of us get "Sora Lite" which will be a heavily nerf'd version.
Again, that's just speculation on my part.
@@TheoreticallyMedia I have a saying I use a lot...that if you wanna predict anything, or solve any mystery...all you have to do is "follow the money." So with that in mind, your theory actually makes Open Ai the most money soooo...Yeah, I'm with ya.
"gonna be so expensive" nowadays its like a lot of ai tools to generously try for free and if it's something you like you can pay like $20/month. All intermediaries are cut out with endless distribution channels. It isn't that bad..
Tim, how do you keep track of all these AI tools? Do you have a master (or mistress) spreadsheet somewhere? A magic 8 ball? I sign up for some of the ones you mention, and by the time they get back to me I have no memory of what they do! Anyway, bless you, you're keeping my up-to-date on all these image/video/audio/text ai's!
Thank you so much! Y'know, it's a really ramshackle system involving bookmarks and notes-- and just constantly scanning for interesting stuff. What I need to do (soon) is put everything together in one place. Basically, everything I've covered on the channel-- into one resource.
Mostly, because I keep forgetting what I've made videos for! Haha-
@@TheoreticallyMedia I had a similar problem with my wordpress blog. I write about lots of organizations, and I was losing track of who I was covering and how often I covered them! In WP, I use tags with the name of each organization, so I can search the blog and see what I've written about them. Then I created a simple text page listing each of the organizations and details about them -- I use that around once a week for checking each organization to make sure I stay up to date.
I know youtube has keywords and tags, although I can't tell what the difference is between them (keywords for channel, tags for specific videos?). Maybe the tags can list each AI program and company for each video. And maybe there's a way to search your channel for videos with specific tags? Seems like a simpler system than creating a database and having to enter all the info into that!
Love that at 3:06! I was tryna catch that fire prompt but your head was in the way 😂thanks for all your research fam!
Gotchu!! Prompt: Soulful Boom Bap Hip Hop instrumental, Solemn effected Piano, SP-1200, low-key swing drums, sine wave bass, Characterful, Peaceful, Interesting, well-arranged composition, 90 BPM
@@TheoreticallyMedia ✊🏽✊🏽✊🏽
Until we see IMG2Video outputs its hard to ascertain how viable SORA really is. Loving Haiper btw.
Just made a 5 track EP with Suno in my first four days of using it! Truly amazing, I'm so proud of what I've been able to produce with it.
Brilliant as always Tim. Thank you. BTW, you say DALL-E is playing catch up with the brush edit. Which others already have it please?
I love how you knocked out a video to compare to the Sora piece. Well done!🎉
I come for the humor and jokes
You stay for the free AI Generated Coffee!
Thanks for the vid yo!
Im actually so glad that any video speaking about SORA on youtube just gets blasted in the comments (nothing toward you, of course :)) OpenAI really is becoming the bogyman of AI. So sick of them gatekeeping everything for hype.
Oh no, I invite it. Legitimately, no flex here-- but I think I'm one of the larger YT channels that covers AI Video fairly in depth. To that, I actually feel a pretty strong responsibility to have a healthy measure of criticism and skepticism with Sora. I've been trying to gauge where you all are on it. We were at around 50/50 a few weeks ago, but as the Sora hype cycle has past its crest, it feels like we're more at like 30/70.
I might run another poll...I'm curious.
Edit: Id like to add too that i 100% agree. I actually had to purge a TON of creators from my sub list as they all fell victim to the hypetrain/clickbait overload.
One of the reasons I genuinely enjoy your content is the way you keep it real. Keep being you, good sir.
@@TheoreticallyMedia Thats actually really interesting to know, thanks for that!
I wonder how much that interview a week or so ago affected it? SUPER bad for optics IMO.
Yeah any comment is a + for the algorithm too !!
Omg I laughed so hard @ the coffee technology!!
I’ll give Sam Altman his 7 trillion if he can make coffee magically appear in my hands!
Haha! Me too!
"Even a slow week in A.I is still moving below light speed" Truer words have never been spoken. I hope a big company brands that statement
Haha, if they do, I'm first in line to sue 'em! Or...maybe just get some free GPU hours from them! I'll probably just settle for them buying me some coffee...haha.
@@TheoreticallyMedia Chat S.U.E
Hey, amazing video man
I wanted to ask, if you don't use DallE, what image gen models do you routinely use?
Do you use a local one or...?
I mostly concentrate on the platform here, since that is the widest/lowest barrier of entry for most people. My two "go tos" are always Midjourney and Leonardo.AI. Although, I've been really liking Scenario lately as well!
You can check them out here: ua-cam.com/video/v_FXC0iq1Sk/v-deo.html
Delving into storytelling and video exploration, VideoGPT subtly refines my content, adding depth and professionalism to my creative projects.
Fantastic Friday update! And Higgs = Best logo! imho.
Hey Tim, great video as usual. The TIM50 code isn't working, any ideas?
The coffee cup appeared in my hand...sigh...I miss the times when cheesy was actually pretty darned funny! I'm saying it was funny.
And Margot Robbie, and AI poetry about AI news, and a headbobbing anime girl and Aphex Twin...great video already without the content! 💯
With Sora, they _are_ beginning to feel like they're heading the 'I am iPhone' feel of smartphones. Suno continues to blow my mind. A month or two (human pace) and they'll have the style reference. Good stuff. Thanks Tim. 👍
Great Video!
Thank you!! That’s great to hear! As you can tell via those mic issues, this one was plagued by a lot of stupid tech issues. Glad to hear it all came together in the end though!
Will there also be other such events in the future? Maybe in Europe? I would love to join!
I heard that Sora takes 1 hour to make a 1-minute video. We waited for an hour for the video to generate, and if the generated video is strange, ( which we don't want ) we have to wait for another hour!!
Yeah it's hype with no real substance for the real world. Hollywood can do amazing things too, if you have the millions of bux for it.
Thanks love 🥰
10000%! Always appreciate the support from you!
@@TheoreticallyMedia love the informative and funiiiiiiiiiiii ;) hahahha
Tim, the video Tim50 does not work at Eventbrite for the Las Vegas AI event. Please fix so I can get two tickets. Thx!
Ah, my bad, it is all caps: TIM50
That’s awesome! If you run into any trouble with it, let me know at theoreticallymedia@gmail.com
It’s gonna be a fun party!
How to even get to Dall-e 3? I have the paid GPT but I see no Dall-e option anywhere?
If you ask ChatGPT to make an image, say "Please make me a Photo of a Sunset" then you're using Dall-E 3.
I know, they should also have the option of a separate UI.
when the first sora video teasers came out, what did I say and post on twitter? i took their prompts, generated in leonardo, and got almost identical videos - just not as long. i said then, there's nothing you can do with sora you can't do with what is already out there other than make longer than 3 or 4 second videos. i stand by that still
Hey! I'm a 'please and thank you' kind of guy too! I guess we're top of the list when it comes to picking the crop of humans that will be spared from extinction when the overlords reign.edit: just saw the other similar comment: *gulp*!
Suno is good, but they currently only allow generations to paid subscribers due to 'System Upgrades'. Will check out Stable Audio and am also currently loving Haiper which is great - more so as its also free and the videos are pretty darn good :D
If you catch Suno in the "off hours" (kind of weird times) you can usually get the v3 model. I get it, they have to prioritize GPUs for the paying folks. Still, it's nice they open it for everyone. Lots you can do with all the free tools!
@@TheoreticallyMedia Handy to know about the 'off hours' on Suno - will try to see when they are for me (UK). Definitely agree that there is a lot of use with the free AI tools, may need to use over a few days but definitely great until afford some of the paid subscriptions :D
What is Thank you sensai Tim
Domo arigato!
Another 'slow week' xD
They say it's never gonna be as slow as it is now for the foreseeable future
Haha, I know. I'm just enjoying the short break to catch my breath. I think it's going to be grind time until November really soon. Everything I'm hearing says this summer is going to be mental.
@@TheoreticallyMedia nervous and excited!
Suno is so good... I've already made 12 songs lol
I love it so much. It is probably the most fun you can have in GenAI right now! The quality is UNreal!
@@TheoreticallyMedia it seriously is...I feel like it's the best moment in functional AI since I first tried GPT 3.5. And I work with AI all day everyday. So impressed. We're so close!
No, you wrote a prompt that had a computer somewhere generate 12 "songs" That's not creation or proof of any talent whatsoever. What are are "learning" how to do has no commercial or artistic value. Its' like someone ordering at a gourmet restaurant calling themselves a chef, and saying they "made" the meal.
@@robertruffo2134 you sound fun at parties. Enjoy the future. You're not going to be happy with that attitude
Thanks for all your work on these videos. You make so many of them and they are of high quality. one thing i think you consistently gloss over with the phrase "think of theses AI Apps as a collaborator". This is a partially misleading statement for my use case. I'm a screenwriter and would love to use some combination of these apps to make a storyboard or even a complete story from my scripts. I find it very difficult to use these apps for anything other than a random shot here or there where you are okay with something 70% of what you wanted. The impression that this channel and many other give is you can make your comic book right now..., I don't think any of this AI stuff is ready so every two months I get my hopes up only to push through a project and get nothing cohesive. maybe you could do a specific course with a set amount of tools and walk us through the process? All of these new AI apps popping up is interesting but they all seem to have the same faults when you want a consistent character with consistent clothing..., ok I feel better now thanks!
Please, thank you ; )
I do sometimes forget to say thank you! But always start with please!
For me DE3 creates astonishing images, that always surprise me,
but the quality is not perfect.
I've seen some great outputs from it-- For some reason, I just tend to struggle with it. I just don't jive with it quite the same way I do with the other major image models. Might just need to spend more time with it.
Re Sora: I wonder if the reason it’s not out yet would be obvious if we knew how many KWh per second of video it burned.
Oh, the compute is stupid high for Sora. That’s why I keep saying if/when it goes out to the public, we’ll probably be seeing the typical 4 second outputs of most other AI Video.
I’ve heard reports that a one minute video can take up to an hour to render as well.
Ouch! that much..
Yeah, I also want an AI that can generate a cup of coffee into my hand. 😇 That equals abundance in my world.
Haha, I'd be 3x as productive! Well, until the coffee crash. 4pm naps would become pretty mandatory in my world!
Tim, Can you please tell us all why SUNO is so good?? Why are they a king? What are they doing differently?
Slow week on _generative_ AI:s maybe, but the big thing right now is Tesla's fsd. It's quite a bit from the theme of this channel, of course, but still.
Haha, yeah I try not to go too far off theme! I keep an eye on all of it though, and yeah, the Tesla thing is pretty big!
Who’s the king of Sound Effects? (Not music) looking for sound effects generator that does an amazing job
Currently, I'd call it Stable Audio-- ALTHOUGH, ElevenLabs is about to hop in the ring. Their SFX model is about to hit beta, and I'm sure it's gonna fly. There's a demo on the home page.
@@TheoreticallyMedia THANK YOU!
🔥🔥🐐🐐😎💯💯💯
The ONLY SORA news I wanna hear is it's release
👋
Heya Louis!!
KEWL
were you first? You get the chicken dinner!!
@@TheoreticallyMedia Shake & Bake Baby!
I say please!!!
We're the nice ones that the AI Overlords will spare! haha
@@TheoreticallyMedia yes!! XD
I know u are just adding the other vid models cause u "have to" but tbh quick math edit 4 secs at the time is 5 times harder than edit 2 minutes at the time
Oh, no-- this wasn't sponsored by Haiper or anything, I just like them. And totally, the 4 second (current) limit requires a lot more elbow grease. More generations, probably playing around with the speed of the video, and playing around with post FX. But, my overall point: It CAN be done. It just takes a little more work.
(Minus those 1-minute long tracking shots. That's a Sora exclusive. For now, at least...)
this is my custom GPT "awesome-story-writer" search , its generating story and images. also I have upload a small tutorial
Wait, where is it?
if you have chat GPT plus, please search in stote Awesome Story Writer...I have a tutorial video with the link at my videos . @@TheoreticallyMedia
you can search ChatGPT plus Awesome Story Write @@TheoreticallyMedia watch?v=pRY_1b_yJrY video link for tutorail, you can add any style number, from 1-450 ,style 1 is cinematic .. style 445 is pixar kind of type
Fuck Sora, they went to “Hollywood” to test - while forgetting the true creators and enthusiasts- and by the time it’s public it will cost $100 USD per 5 second HD render
Playing devil's advocate, it could be a hardware issue. Let's say running SORA requires an A100 and takes 1hr to render 1min of video - if you're were them would you release it for free, or try and get people who can actually afford a $10k GPU to pay for it and help fund development?
the only interesting thing about chgpt dallie, is its ability to recreate light resembling a REAL light source.
Yeah, that illumination on the toast was really nice, but other than that, I just find Dalle to be pretty uninspiring. Shame really. Maybe they’ll pull a rabbit out of their hat for v4
can you generate nude models or are they all censored?
On that subject . Midjourney is generating topless models , unprompted. It will censor a requested prompt then generate an unrequested nude ? That’s without using any clever prompts .
In Dalle? No way. You can barely generate someone that looks photorealistic.
They’ve been playing around with that a bit. I think they’re trying to figure out how to allow it for artistic purposes, while not allowing gratuitous nudity.
It’s a tight wire rope for sure. I’ve had to do some MPAA type classifications on movies like Superbad. Trying to judge if a F word is for comedic purposes or not, I’ll tell you, it’s a tough gig!
FOR FUCK'S SAKE, OpenAI, just give us a DALL-E 3 Labs site already. The Labs site for DALL-E 2 was the best/simplest/most elegant image AI UX out there, it just needed an engine upgrade, which they have so far confoundingly refused to give us (even removing the previous, already belated promise of an eventual Labs site from their official DALL-E 3 page).
It never made a lick of sense to me to position ChatGPT between the user and DALL-E 3. There has never been any significant advantage in having ANOTHER AI translate your text prompts into... slightly different text prompts. Plenty of disadvantages, though, and way more opportunities for misunderstanding.
Haha, I don’t mind the interface between an LLM and an image generator, I think if it had some personality like Claude it could even be fun. But, agreed: the old interface was great- and honestly, ChatGpt just gets in the way.
There’s also a stupid amount of guardrails, which I didn’t address in this video, but I’m not even trying to make anything remotely controversial in these videos and was getting hit with it.
Hence, Toast.
Nothing against toast ;) But I am much less of a fan of ChatGPT in general than I am of most of the image-oriented AI's. I did manage to get it to write a cover letter for my resume, but other than that, my exchanges with it (especially when attempting to get it to make images) are generally comprised of a series of reasonably eloquent apologies for failing to do what I asked it to, followed immediately by more failing and more apologies, ad nauseum.
Is higgsfield beta vs apple app available different anyone know ?
They mentioned they had an app in the interview, but apparently not available in the US? Maybe that changed?
Slow week hence he is talking faster
Haha it was all that AI coffee!
Create me... is just weird.
If they don't plan to launch the model quickly (like Sora), then why do they entice us by talking about it so early!!!? 🥴
I would not be surprised if SORA doesn't launch until 2025...and by then no one will care and or be using another Video Ai tool.
oh, I 100% agree that they're going to be eclipsed by another model before they even release. And I don't think we'll ever see the full 1-minute generations in Sora. At least for awhile. That burns WAYYY too much GPU to produce.
LLMs are trained on text written by humans. When you say "please" to a human, you're more likely to receive a helpful response. Therefore it seems perfectly reasonable to me that saying "please" to an LLM could yield better results. If chatbots are purposefully made to be human-like, then why be disrespectful? Seems like a bad habit to form.
AI companies need to be better at their names. OpenAI aren’t “open” and now stability AI isn’t stable :P
I don’t find Midjourney mid either!
i dont know why you guys are promoting a lame a8 that hasnt been available to public yet.
I'm not promoting it. In fact, if anything I'm a bit critical of Sora. There was a whole other side rant in this video about it that I cut out...mostly because it was too "rant-y"-- no one wants to hear that!
If you just yhink a while instesd of breathlessly be hyped by all AI toys you would know why SORA is not public yet.
Until it is safe to use and we are able to know what is AI or not SORA should not be unlesshed. I know that others will reach that same point soon but unless you are willing to live in a world where we cant tell true from false just so you can play and distract yourself…then I hope you also can take the blame if it goes south.
I mean, look-- we already can't. And we haven't for awhile-- Even going back before Tom Cruise Deepfakes or whatever, any VFX artist could create a pretty convincing scene of just about anything. It would just take them a lot longer to do.
I could go into a much longer diatribe of the post-truth world, but that's not what this channel is about. So, I'll just say: Since it's already here, why not release it to the artists who might make some positive stuff with it for a change?
@@TheoreticallyMedia I agree in much of it and love to create myself. But whatever time we can get to delay a possible complete crash of democracy and the free word is worth the wait.
In the end only those that already have the money and the power and wants to keep us in conttol will be able to manipulate what ever ”reality” we have left. I also want to do amazing stories in SORA but am also thinking that maybe we should think ahesd this time.
Propaganda made in AI might be easy or possible for us in the rich world to see through, but then the majority of the people of Earth are not lucky to be that well informed and well educated as we are. This can lesd to conflicts and wars or religious sect like situations that affects is too through trade issues and terror acts among many dsngerous side effects.
I can see what MAGA has done to the USA with normal social media and a person, Donald Trump, that claims Godlike worship from all around him. Just think of how much more confusing/convincing these type of despots would be with enourmous compute capability.
I know there must be a balance. I am not against AI. I am just against how it can be used to do harm and turn a possible creative freedom into a horrible control tool where you, me and everyone not in poeer are not allowed to express ourselves freely.
Thank you for reading and answering! Respect that alot. 👍
Cheers!
Regarding Sora, this latest batch didn't impress as much. I'm getting bored of psychedelic abstract flights through the looking glass.
Good luck with judging the eSports stuff and getting away with your casino heist!
FWIW: Tim50 code doesn't appear to be working. You may want to double-check it.
I messed up: it’s all caps: TIM50
I’ll change it in the description!