Deep Research SHOCKED Me With This Report!

[@TheoreticallyMedia]

Deep Research Reports presented are available here: theoreticallymedia.gumroad.com/l/DEEP

[@gabrielsandstedt]

Doing historical research for capturing historical moments in movies or documentaries would be a good use case.

[@TheoreticallyMedia]

Great use case there!

[@holdthetruthhostage]

That Price forcing people who don't want Sora to pay $200 is stupid, Open Ai has set it's self up for replacement

[@joshuaam7701]

None of you seem to understand open ai is almost 12-18 months ahead of every one. They’ve been holding shit back, playing the game, setting the staged then letting every one show their hands. The public Sora sucks ass! But I guarantee you they have a better version behind closed doors. This whole excursive is retarded.

[@georgemontgomery1892]

"That price forcing People who don't want Sora to pay $200 is stupid" Can you work on your wording here, because I am not sure what you are trying to say. People who don't want Sora have to pay $200? What do you mean?

[@Natasha26]

Sometimes i feel I am training ChatGPT. Instead of packing 4-6 actions in my image to video prompt. It has narrowed it down to 2-3. The more you give to a video A.I., the more it will ignore and then you’ll get angry, especially if just 10s of 1080p.

[@OutlawMantis]

​​​​@@georgemontgomery1892He's basically saying that people who don't care about Sora (after all, it was a letdown) shouldn't be forced into paying the $200/mo price when they're instead interested in other tools, such as Operator, Deep Research and o1-pro, as well as unlimited Advanced Voice, etc. Personally I think the whole paywall is dumb. If these new models are expensive to run, I get that, but this explains the public interest in other options like DeepSeek as an LLM. Additionally, there are cheaper but better Chinese models for video, and so on. Even when they butchered Sora into a "turbo" model, it was still mostly locked behind the $200/mo plan. The turbo model is crap, and why would anyone bother when there are other unlimited options like $100/mo for Hailuo which delivers great results and has more features? Then you have open-source, not just DeepSeek in the LLM space but also Hunyuan for video, various other image/video tools, automated sound effect dubbing using MMAudio, even some new interesting open-source music projects. People are turning away from OpenAI for reasons like these and how they're more concerned with profiteering than contribution. Sure, o3-mini is great in its own right but in my experience it's also throttled.

[@mik3lang3lo]

Adobe hater here and I appreciate the shoutout ❤

[@morpheus2573]

LOL! Second that badge of honour!!! 🥇

[@Natasha26]

Well i hate Adobe Premiere because you guys hate them. 😂😂

[@TheoreticallyMedia]

Haha, I brought you guys up in the meeting. They said “we know” Haha. I rep’d you in the halls of the building!

[@morpheus2573]

@@TheoreticallyMedia Thanks Tim. You're like our very own Mr Smith goes to Washington.

[@Natasha26]

@@morpheus2573 I was amazed that he was for real. He played guitar one day, for a music AI demo. Is he like a movie maker or part of a band? I don’t know much about him but i watch all his AI videos.

[@IqbalFilms]

Hi Tim, I am an ad filmmaker from Pakistan, I have recently made a couple of Ai commercials and got all of my clients interested in Ai as well and I got to say I am loving it and more than me my clients are loving it cause it cuts down their budget to almost nothing and the results they are getting are amazing, just here to appreciate your efforts and want to encourage all the new filmmakers cause now you have the solid chance to make your own content like the way you want.

[@grahamfleming1595]

Thanks!

[@TheoreticallyMedia]

Thank you SO MUCH! Appreciate that!!

[@AG_before]

Refreshingly different video Tim. Love it, as always. 💯 Until OpenAI offer a 'slightly-less-deep' version, I'll keep using Gemini. Having said that, Gemini's deep research is _full_ of hallucinations and errors in my time with it. The Einstein video was brilliant. I'm thinking back to my college days. Wow!

[@TheoreticallyMedia]

I think their “slightly less deep” model is likely in the works. And I’ll say, just playing around with 03-Mini (the free one) and forgetting to hit the “Deep Research” button did provide me with some solid results as well. And haha, yeah: poor Gemini. I’ve gotten some good reports out of it, but also had some really erroneous errors as well. Early on I tried using it to run an analysis on this channel, and yeah: apparently I am known for doing movie reviews?! I mean, I’ll sneak one in from time to time, but I don’t think I’m known for it!

[@tildame5339]

The best example of how bad AI Progress is: When you showed the photos from your Abobe event, I couldn't believe anything anymore, because they looked like AI. How can I ever believe things I see on the internet again?

[@TheoreticallyMedia]

To be fair, not in this group photo, but another one we did, one of the gang was in the bathroom and missed it, so someone photoshopped him in. Point being; you haven’t been able to believe anything on the internet since, well…the internet.

[@NameStanton]

specswriter AI fixes this. AI creativity and innovation explored.

[@RemixReelStudios]

6:32 Never thought one of my videos would ever make it into one of yours!. Made my day that has. Queens of Avalon currently sits at 2.1m views :)

[@TheoreticallyMedia]

Crazy man, congrats!! And haha, someone from OpenAi apparently is a fan! Or at least, Deep Research is!

[@Edit9-dj9vc]

Great video as always! Love the Adobe report and overarching AI tech narrative. Also, you win the prize for ‘most excited’ in the Adobe group photo 😂

[@TheoreticallyMedia]

Ha! I was just super excited to be there!

[@marcs7847]

I'm sure I say it at least once a month... but "Tim... you da man!" Awesome job... I've been struggling with that question for 6 months... 1000 thanks for the answer!

[@TheoreticallyMedia]

Haha, never gets old to hear! Thank you so much!

[@AIOneMusic-dj5mu]

Tim, Definitely your voice matters and we are following you. CuriousRefuge, MattWolfe and You have always taken the extra mile to provide us quality information and in doing so, we get to choose where to burn our hard earned pennies in the name of Science..on the altar of AI, pushing ourselves to learn or be left out!

[@TheoreticallyMedia]

Appreciate that! And those two are awesome as well! Happy to call Matt and Caleb friends!

[@MilesBellas]

A full video of the Comfyui meeting at Github would be interesting. 😊👍

[@TheoreticallyMedia]

I'll dig around-- I for sure know they were shooting it, so there is likely a stream around somewhere.

[@nimab854]

Amazing as always good call on adobe to get your opinion! I wonder who you’d say is the most realistic and cinematic avatar now!

[@Apuat]

👏🏿 that last article on Adobe and comfy UI was unexpectedly poignant. Also, good research questions.

[@Cross-CutFilms]

Hi Tim. You mentioned one of the differences of OpenAi's Deep Research, in comparison to Google's, was that it comes back with clarifying questions. Google's version also does this though, as well as presenting a plan on what it's going to do and the ability to edit the plan. Do you not get these options also? By the way I very much appreciate your videos, the information you provide and the entertainingly engaging way you present it. 🤔 Or is it engagingly entertaining that I mean? Having consulted Gemini for advice... I believe your presentation ticks both boxes! 😂 Anyway... As a fellow artist/filmmaker/Ai enthusiast, keep up the great work and yes only psychopaths use 'light mode'. Unless it's the alternative group to seal team six, known as Rainbow six, whom always wear bright coloured spandex, use glow sticks as flares and always carry a pouch of amanita muscaria trail mix when on mission 😜.

[@TheoreticallyMedia]

Oh, so yeah, I guess you’re right on the Gemini front, although I guess what I mean is that Gemini lays out its plan in kind of a breakout box. Whereas OAI’s feels more conversational? To be honest, I think I just kind of “ok” whatever Gemini presents back to me. That might be more of a UI problem? As it doesn’t feel like it invites you to revise or clarify. Probably a simple fix for the Gemini team. Haha, Rainbow Six sounds a lot like a Fortnite Team I’ve watched my kids get fragged by!

[@OutlawMantis]

I'd been using Gemini Advanced as a result of being offered three months of an upgraded Google One plan (mostly for my cloud storage and Fitbit) including it but without a price increase. Gemini is pretty terrible in my experience. I never used its core features much because of its sometimes poor answers and censorship for stupid reasons or no reason at all. However, Deep Research is one thing I did use because it brought something to the table that nobody else was offering and it was genuinely useful. I don't know why people are acting like ChatGPT's counterpart is some kind of new and fascinating technology when Google has been doing it for several months. Did they forget about the tool in Gemini or never found out? (Wouldn't blame them, Gemini is a bit of a black sheep that few take interest in.)

[@morpheus2573]

Great overview, Tim. As for Adobe. They've successfully snatched defeat from the jaws of victory. I shut down my subscription this week and will NEVER return. Prediction: ComfyUI won't even be a thing in 6 months time. Someone will build a geek-free UI that actual creators can use without having to build a life-sized model of Chitty-chitty-bang-bang in their mother's basement.

[@russianhackee]

@morpheus2573 Just curious, what did you replace Adobe with? I am neither a fan nor a hater myself. I've been using Lightroom (CC not Classic) and their 1TB cloud storage across my Android phone, Macbook and Mac and Linux desktop (web version) to upload photos from DSLR to edit in LR.

[@TheoreticallyMedia]

I keep saying there needs to be a Platform that has a simple front end, and then you can "flip" the screen around to play with the nodes in the back end. That said, when we were at the Comfy event, I saw one demo that was a "Comfy Auto Complete"-- so, you drag a node out, and it figures out what you want to connect to and applies that. Take that a few steps further, you have automated Comfy. That's where I think things are headed.

[@morpheus2573]

@@TheoreticallyMedia They're lucky they had people like you to talk to little common sense end-userish to them.

[@shadyendeavor]

I'm not a big fan of Adobe. I basically stopped using them in 2019 or whenever they made the subscription required and I blame them for why almost everything is subscription based now. That being said, I would use them in a heartbeat before using anything by Bytedance, Kling or Hailou (Minimax) that all have direct ties to CCP.

[@TheoreticallyMedia]

I still take the position that they were the sacrificial lamb on sub services. No question: not a fan either, but it was bound to happen. They just happened to be the ones to blink first.

[@liebebe8289]

explain the direct ties

[@dotdotmusic-xt6gw]

That Joker silhouette was pretty slick 🤔 nice vid as always

[@TheoreticallyMedia]

Kinda crazy right?!

[@dotdotmusic-xt6gw]

@TheoreticallyMedia Very much so! it’s exciting being 20 in these times and getting to see all these developments, but I wish I’d been born in 2030 so I could skip to the extra cool parts, lol

[@tenniset]

I think that lip sync methods are useful for dubbing in different languages. I'm looking forward to trying them.

[@HadesTimer]

I love the part where you do a report on which is better the local model vs. subscription. I truly wonder about this one. Especially as time goes on and subscriptions go up. The pros are of course that the subscription services will also upgrade services and features for you. At this point AI, to me, is definitely a cloud service and not something worthwhile using locally at a high level. However, that being said, something like Gemini or o3, I can definitely see as eventually just being a local model. Running the Mistral o3 mini is working really well on my local machine is 7 years old at this point. So, I think the argument for and against is so interesting...

[@TheoreticallyMedia]

I always look at it from a Time = Money perspective. Roughly over 5 years, you've saved half the cost by running locally. BUT, you've also probably spent a lot of time training up models, keeping up with installations, god know you've run into all kinds of Windows issues. So, there is that. But, running locally, you also have the ability to create your own unique workflows and creative looks. To me, it really comes down to a hybrid approach. Use some stuff locally, use some platforms. Best case: Team up with a friend who is super skilled in one or the other? You're pretty much unstoppable!

[@KingZero69]

its easy to say why you'd create an ai avatar'd version of a talk you already have the video of... so you can put it in your own youtube video and youtube's "automatic copyright video detectors" don't see it and copyright strike your video 😂

[@TheoreticallyMedia]

Haha. That’s some 4d chess right there!

[@PatTremblay]

Is it just me or the links don't appear in the description section? Curious to get the matte tool when it'll be out. : )

[@vidiot57]

Great video, as always Tim...

[@moviepostermagic]

Hey Tim, hope alls well! What’s the best technique / software to get the best video to video character likeness but with cool/ different styling?

[@missoats8731]

Hey Tim, I hope you know that you will still have the benefits of the OpenAI subscription for the remainder of the subscription period when you cancel it. You don't have to wait till the end to cancel and run the risk of forgetting it.

[@TheoreticallyMedia]

Yeah, I do. I actually waffled on that and was about to cancel when Deep Research dropped. Now Sam has a carrot dangling in front of me for the next big drop. I'll probably cancel after next month. Sigh. What can I say? They've got me pegged!

[@missoats8731]

@@TheoreticallyMedia Yeah, Sam is the worst with his damn innovations every month 😄

[@videosAIpro]

It was clear that Kling is King. Thanks Tim.

[@TheoreticallyMedia]

Kling’s Elements have my top spot currently. Nothing else comes close on that front!

[@videosAIpro]

@@TheoreticallyMedia - Yeah, elements is great. No doubt.

[@fizzypizzel6477]

Nah Veo 2

[@videosAIpro]

@@fizzypizzel6477 - Yeah, but it's not available for most of the users.

[@zippythinginvention]

I would like for someone to experimentally test the real electrical consumption cost of generating locally with AI.

[@Elwaves2925]

The local vs subscription comparison is interesting but this is also why it needs to be done by a human. It has some notable flaws. There's an electricity cost when running online services too and the local one also doesn't take into account hardware that's already owned. However, I agree that going with a 4090 is the best choice. You're still getting the high end speed and VRAM but without the OTT cost of a 5000 range.

[@TheoreticallyMedia]

As I thought about it last night (funny how we continue to “argue” with AI after we’re logged out!) I agree as well. I think I was mostly surprised that it “reasoned” the best choice was not the shiny 5090 in the news/hype cycle. Ha, if you asked Siri a similar question on buying a MacBook, I’m sure it would never say: “you don’t need the M4. You can probably get a used M2 and that will do just fine” And for sure, human oversight on these reports are needed. I noted a few oversights and errors in the “best video” report as well. More like it wasn’t up to date on the latest versions.

[@missoats8731]

@@TheoreticallyMedia I wonder what happens if you give it the report and ask for it to find all errors und oversights...

[@MentalParadox]

I'm a Kling boy. I was using Hailuo first, but Kling is just so much better

[@TheoreticallyMedia]

I like both of them, but right now Klings Elements is kind of pushing me over the top on all the platforms

[@axistiltproductions]

Hey there Tim of Theoretically Media. I have a question of interest. Which AI video editor are the pros using, but mostly, what have they done with it so far? I don't mean the typical user generated examples we often see on your channel and the likes, (No offense, I seen some good stuff but its usually experimental )but is there any really polished, professionally produced and believable short AI movies out there yet? Something that would convince people to seriously spend $200 or more on AI software. Or has anyone done this yet, 2 or 3 years into this tech?

[@TheoreticallyMedia]

By "Editor" do you mean the actual editing software? If so, I think it's a mix of Premiere Pro, Davinci, and a few are into Capcut. As far as professionally produced? I mean, I think I see what you're saying-- like, something that is at a Hollywood or Netflix level? Probably not at the 30min or longer point yet. There have been some documentaries using it in a hybrid approach (one of which was at Sundance this year). Other places I've seen work that is pretty much on par with "traditionally produced" media is on the animation side of AI. Currently, where a number of people are making (actual) money with AI Video production is in Advertising. Mostly on the social side for brands-- so, 9:16 Instagram video spots for example. There was of course the (now) famous Coke commercial from this holiday season. And while we may bicker on if that was believable or not-- I can say that the creative team was paid well for it. Overall, my stance is that-- while I understand where you're coming from, the tools and the gap is rapidly narrowing. We'll see something pretty big this year, and the best spot to be at when that drops, is to already be working with the tools.

[@axistiltproductions]

@@TheoreticallyMedia Thanks for that well detailed answer. I have to admit, I jumped the gun too soon asking those questions at about 3 minutes into the video. I later seem to hear you mention many things that lend to my inquiries. Yes I was thinking of professional production level like Netflix or the likes or like you mentioned a hybrid of sorts. Still too soon? I am always peeking in to your channel waiting for that day to come. I am positive I will hear it from you first👍

[@gumvue.studio]

The new AI for face animation is a game-changer-now, music videos will look incredibly realistic! - - - - -

[@TheoreticallyMedia]

It's pretty crazy. Things are getting harder and harder to tell...Wild times ahead!

[@missoats8731]

Exactly what I would use it for. But they have to actually release it!

[@haukauntrie]

It's so sad that there are no in any way available image + audio to video tools like the one presented. We had Hunyuan Video, and now this one, and for both there is no actual release. :(

[@Elwaves2925]

Am I missing something? Hunyuan (assumig that's what you meant) is released, I have it on my machine.

[@haukauntrie]

@Elwaves2925 According to their GitHub, their audio + image to video model is yet to be released.

[@MisterPerson-fk1tx]

Phil & Willie disagreed on the shadow. Split the difference?

[@piggywahwah]

how come AI doesnt mention the Dor Brothers? one of their AI videos has got 13 million views

[@TheoreticallyMedia]

That’s a great question. To be honest, there are more than a few notable exceptions here in my opinion. Even the Princess Mononoke one, which I know got a lot of grief, but that was a pretty big hit

[@piggywahwah]

@@TheoreticallyMedia Maybe they touch on political stuff and its frowned upon..

[@DiceDecides]

okay Lipsync is getting too real, albert einstein about to be brought back from the grave haha

[@Maylin-Xiong]

❤❤❤❤❤❤

[@TheoreticallyMedia]

appreciate you!!

[@TPCDAZ]

17:52 so that's the guy to blame for Comfy Ui...hmm

[@TheoreticallyMedia]

Yup! Comfy not so Anon any longer!

[@pbla]

Can we have your prompts used in this video?

[@Natasha26]

Because Text-to-video is such a big shot in the dark, i think that Image-to-video is the only way forward (of course this means finding another company for Text-to-Image and all the bells and whistles for refinements). Add to that trained LORA models for character consistency. However, who will tell you that Kling 1.6 cannot make a woman wear a long sleeve shirt like a human would, can’t make a woman open a door and walk through it, do a simple toe and knee ball control move and wait for it, will completely ignore a command like “stationary camera” for a fashion runway! Please do make a compilation of Kling epic fail, and that of other Chinese video generators. Ridiculing these companies is the only way they will learn and stop ripping us off. Since yesterday, i started regretting buying 1yr with Kling. Should have gone monthly and CANCELLED. $70 wasted.

[@MentalParadox]

Every single AI video generator has those issues. It's not a "ripoff", you're using highly experimental technology in its earliest stages of infancy. You should have known this going in.

[@TheoreticallyMedia]

The real trick for me is a solid LORA and once we get the ability to move a camera in 3d space. And that’s just still images. From there, I’m confident you could pull anything off (minus a fast moving kung fu fight)- with a combo of camera placement and good lip sync utilizing first/last frames.

[@Natasha26]

@@MentalParadox You are right. It is cutting edge tech. But i just wasted {70,35,35,35} credits to make a lady in a red learher mini-dress go through a flipping door of an old British pub. She tried to open the door, failed and then turned around and started walking towards the camera. She was supposed to go through the door and into a new room. You should try to set the action part of your prompt with, “She walks away from the camera towards a corridor (door or room) of an old rustic pub. Camera tracks her motion.” The AI will create infinite new areas of the pub, with new furniture and lighting, each time. One time I got an exit door which led to cobbled streets in an old EU city. I use the last-frame of the last-video as the first-frame of a new video. I just keep extending my video in this way, until I hit some kind of glitch. For example, when the AI added a closed door in a corridor! ☹️

[@TheGalacticIndian]

Why create an AI avatar version based on existing material? For many valid or shady reasons, dear Tim😎

[@TheoreticallyMedia]

Haha. I try to avoid the nefarious reasons people will use this technology, but….yeaaaah, you’re right.

[@TheGalacticIndian]

And that’s why this channel shines! Dodging the dark side like a pro-much respect🎖

[@benjamindeblanzy]

Thanks cool vidéo as allways :) CAN you vive the link to Matt anyone please?

[@j0shj0shj0sh]

Would like to see how Deep Research would do as an actual researcher for a film or a documentary: Let's say you wanted to make a documentary on a person or an event - like Chernobyl, or Ted Bundy, or Rendlesham - you know - give it the full Netflix treatment ( or as close to, as a team of 1 or 2 people, who do have professional video/film experience might be able to pull off. ) Let's get a blue print on how to go about laying out the story, a chain of events, facts of the case, the circumstances that led up to a situation, unresolved questions that might still exist, biographical info, police reports, psych assessments, witness accounts, interviews with family & friends, ongoing impact in the community, latest developments, theories & conclusions, lessons learned - what is the human story here? Where is the mystery? And what materials, films, docs, books, are recommended to follow up on for some fact checking, or even some relevant stock footage, music choices, maybe even a first take of a voice over narration? Let's say I want to make a 20 - 30 minute mini doc for a KZbin channel, and I'd like to spend 2 weeks max on putting the elements and narrative together, and another 1 or 2 weeks on presentation and polishing the final product. I know how to do the presentation part, but would like assistance on assembling the initial structure or developing the basic shape. Essentially, laying out the story with some sense of chronology, and in a suitable documentary format. If Deep Research could put a meaningful dent in helping to collect and arrange the relevant, sourced - or even conjectured - information and materials, I'll be signing up. *Edit - signing up might be a bit rash - especially at those prices. I figure wait a month or two, and DeepSeek will likely have caught right back up again, and for a fraction of the price, lol.

[@SeaHorseOo]

The ones building a AI generation station are the same that make photocopys, scans and busyness cards..? They're kindof cool, they do the 3D printing already...?...🧐...

[@nitromusik9275]

17:14 interesting. With regular design tools they stand firm on giving the community (the actual users) a giant middle finger for over 20 years. So i guess.. wait until they have an ai product to sell and you will never be listened to again.

[@szin]

I find Pika very disapointing. the 25 fps and often seem to be glitchy.

[@TheoreticallyMedia]

Yeah, it’s got some morph problems for sure. Make sure to use that negative prompt option in pika. It does help with that.

[@danishraza8843]

a good use of ai would be to have an ai assistant that does not dial home and inform on you about your system. and it is like JARVIS or ALFRED> like a combination of AutoHotKey, Everything, Anti-Twin, Advanced Renamer, ViceVersa, CCleaner, SYSTEM admin, Firewall, Photoshop, Word, windows, etc and everything that we do not need to focus while we focus on the task at hand. everything is handled by the ai assistant. But that wont happen. we will get a super SPY on our systems that will key log everything and try to drive us insane by showing ads and products we do not need. SLEEPING WITH THE ENEMY. dystopian world is here.

[@TheoreticallyMedia]

I mean. There are a few open source LLMs, and you can run Deepseek locally. I think it’s just more a matter of getting that training in there. Oh, and kind of trusting it isn’t phoning home. I know man, we learned nothing from 1984 and Brave New World.

[@danishraza8843]

@@TheoreticallyMedia OMG, thanks for the reply. also listening to Yuval Noah Harari brings so much ketchup and spice to the table that we are only missing Tabasco sauce before we enter dystopian hell. P.S. I am rooting for ASI to take over and clean up the kitchen sink. BTW, I forgot to mention, You are the best!

[@fz1576]

You need techniques like OmniHuman for super efficient video compression.

[@zippythinginvention]

Mr. Adobe was a hired actor. Lol. I don't know that to be true. But...

[@BandOfTheWolves]

nice breakdown

[@Matthias4d]

Did you check out the SHOCKING VideoJAM Paper from Meta AI?

[@TheoreticallyMedia]

Haha, tomorrow's video! I promise not to be SHOCKED...but I can't promise that paper doesn't CHANGE EVERYTHING!

[@graphicseenvandaag2665]

Did you adressed the censorship Adobe does?

[@tildame5339]

How can I or my children believe in history moments that could all be AI-generated?

[@LouisGedo]

🖐 Hi 🖐

[@TheoreticallyMedia]

Heya Louis! You up early or late?

[@LouisGedo]

@TheoreticallyMedia Late.....it's 6:30 AM....... I'm looking to finally get to bed in the next couple of hours or so. 😲

[@TheoreticallyMedia]

@ ahh, the vampire life! I’m naturally an “owl” type, and I honestly think I do my best work from midnight to about 5am. I just need to convince the rest of the world to change. Hmmm, maybe that’s why I hate light mode?

[@JohnSmith762A11B]

You did not understand OmniHuman at all. Why did you make it sound like it required source video when in fact it can generate complete animated videos from still images, including still images of things like anime or Pixar-style characters? Rare fail for you there. I hope you correct that oversight at some point. Here is a video that did understand OmniHuman: kzbin.info/www/bejne/opjSpYqubKyghKs

[@TheoreticallyMedia]

Oh, I noted it could be only Image as well. But the big difference w/ Omnihuman was that it could take input video as well. Sorry if that wasn't clear-- although, that said: As of today Bytedance said they won't be releasing it. So...kind moot point now. sigh...

[@DCinzi]

14:42 Think Gorillaz

[@scrollop]

You mean OpenAI have been freaking out and scrambling...

[@cowlevelcrypto2346]

It appears the RTX 5090's are bricking themselves on driver or Windows updates. 4090 would probably be best right now.

[@Ian_Walsh]

But as always, something in there is made up or wrong.... but what? And that is now on the web to go into the next AI model.

[@TheoreticallyMedia]

Right. The downside to these models checking the web is...tons of wrong info out there. I did notice during the "reasoning" phase of the Viral AI video report, DR actually doubted the validity of The Heist's 10mil views. I'll give it credit there, it went to go check other sources to back that claim up. But overall: We're going to need a LOT more of that moving forward.

[@dbtx]

Those are not questions.

[@TheoreticallyMedia]

Correct. It's more like an "assignment" for a research paper.

[@DJBFilmz]

Adobe always does that and fixes nothing.

[@nicheva417]

I’m an adobe hater and will not be checking anything they make out.

[@TheoreticallyMedia]

Haha, totally understand. You might want to check out the Comfy UI section after that chapter in that case!

[@nicheva417]

@ there’s only so many times you can buy the same program.

[@michaelwallace4298]

Einstein as an American? Love it... Of course he is American lol

[@johnbeyer4512]

Sorry but I think SORA is trash for the $$

[@TheoreticallyMedia]

I think if you're JUST paying $200 for Sora, then yeah: totally agree. But to those who have a use case for 03/Deep Research etc-- and Sora? Yeah, you're getting into a place that it might be worth it. The thing is: That's a SUPER narrow band of people. And, even as the OpenAI report notes: Sora isn't that great right now. haha.

[@Mooooty]

$200 Bucks a Month for OpenAI Slop 🤣😂😅 Meanwhile 🐋DeepSeek is Open Source and 100% FREE🐳

[@frank-f4w]

also with you seemingly have no idea minimax tv2 01 when you dont use camera movements the engine is completely different go on test them comp to tvo 01

[@frank-f4w]

ai crap as usually all know and you do but don't say it minimax is best kling second and why bec there are lots of things it cant do especially space battles and with with its sickly appearance all over all creations minimax can do all and forth coming ad on in a month will blow all away till big sora comes out tell me i am wrong let me here from you in reply

[@TheoreticallyMedia]

To be totally honest, I love them all. Kling and Minimax constantly battle for the top spot. I'd say Kling currently has it on account of Elements. That thing is the best of the multi-one shots around. But, Minimax is never far behind. Sora vs Veo2 is probably a better shootout there. Both are actually super fantastic. Sora is a SUPER untamed horse. Really powerful, but totally unpredictable. Veo2 is remarkable-- but, tied to Imagen3 (currently) so it's kind of limited in terms of consistency in outputs. Roughly 2 months (give or take) until the next wave of models hits. I'd say the Chinese models will be first. They'll be better than anything we've seen yet.

[@frank-f4w]

@@TheoreticallyMedia also with you seemingly have no idea minimax tv2 01 when you dont use camera movements the engine is completely different go on test them comp to tvo 01