← AI For Humans: Weekly AI News, Tools & Trends
AI For Humans: Weekly AI News, Tools & Trends
Seedance 2.5 & The New Age of AI Video (w/ Theoretically Media)
00:53:25
Loading summary
Transcript212 lines
- [00:00]
A
The next generation of AI video is here with Sedance 2.5, Minimax H3, Wand 3.0 and Flux 3. But maybe more importantly, Minimax's new model runs on local hardware and will let you do things like this. Walter. But nearly all of these models are Chinese. Does that mean that America has given up on the AI Video frontier? Yes, Gavin. Also, haha, funny innuendo comment. What a weird human you are. Get behind me, A.I.
- [00:42]
B
kevin.
- [00:43]
A
Yes, Kevin's on vacation today. But thankfully we are joined by incredible AI Video youtuber Tim Simmons from Theoretically Video.
- [00:49]
B
Oh, this again.
- [00:51]
A
Yes, Tim, this again. This is AI For Humans. Welcome everybody, to AI For Human. Humans. We are very excited to be joined today by a guest host. Tim, welcome. Thank you so much for being here.
- [01:09]
B
Absolutely. Breaking news, by the way. I know Kevin's on vacation this week. Next week I'm on vacation.
- [01:15]
A
Oh, well, this is perfect then, because you'll get your double dose of Tim. So if you're coming from Tim's channel to watch us, maybe, you know, watch it today or save it for next week. One way or the other, you got to get your Tim.
- [01:26]
B
The news will be stale by then, but I will be fresh.
- [01:29]
A
You're right. Okay, so today I'm very excited because we're having you on. This is the perfect time to have you on. This is another giant moment in AI Video. We've been talking about this in our show for a while. I know you've been talking about it on your channel. We are entering into what I would. I guess it's the next generation of AI video. We have CDNs 2.5 today, live for all of the world, including America. I will talk about that, my experience with that in a little bit. We have minimax H3, which there's a lot of interesting stuff being done literally right before we start recording wand 3.0 and then flux 3. So Tim, I know you are an expert on this. If you are not following Tim, you should definitely go follow him. You've used most of these except for 1 3. Before we even jump into the details of any of these. Like which one are you thinking is kind of state of the art right now?
- [02:15]
B
For. Just real quick, just for historical context, because I always have to mention it as 13 has now dropped. When Wan first dropped, when the initial version of it, it was called Wan X or wank, if you will. So I will never. They quickly changed the name. I will never let them forget that. Yeah, so I've used all of them. But. But when. I mean, I still think that 2.5 is an interesting beast because it is good in some ways, but it is a step back from 2.0 in other ways. I think that Hylou Minimax is H3. That is very good. I would say that it is probably just like a beat step Behind Sea Dance 2.0, but it can be run locally and I have been running it locally like crazy on my PC, which is sitting over there. That's wild. And then Flux is. Flux is in the race. It's funny, I haven't really played too much with Flux since launch just because that sort of onslaught of stuff just started happening. So I haven't gotten Circle.
- [03:21]
A
Well, it's funny. They all came out kind of at the same time, right. And it is tricky. Well, let's, let's. Okay, let's go through each of these individually because they're each kind of big news today. So today let's start with CDNs 2.5. This is. I think Sea Dance 2 was my kind of like, holy crap moment in AI video. I've been doing a lot of work in Sea Dance 2 over the last three or four months. I know you have as well. The ability to get like really good results out of it pretty often was much higher than it has in the past. 2.5 is. The biggest differences here are you can get 30 second outputs, you can put in a lot of. Into the Omni model, you can put a lot of references.
- [04:00]
B
I think it's like 50. 50, 50. Yeah. Y.
- [04:03]
A
Which is crazy. Now I will say my experience with this so far because unlike some people, I am in the dream meet a ccp. But I did not get the early access and I have been trying something today and I want to see if one of them has resulted. I am still. I. So I have been trying to do a video and I've gotten rejected most times. I've got three that are stuck at 99 right now. And what it was is I took my image, my audio, and then I took the entire male product collection. I just took screenshots of it and I basically set myself up to do a fashion model shoot. I have been rejected a lot here, Tim. And I don't think it's because I'm not model material. I think it's because there are still significant things you cannot generate in this model. Has that been your experience? Not too.
- [04:48]
B
I haven't run into that too much, but a lot of stuff.
- [04:50]
A
Yeah.
- [04:51]
B
A lot of the stuff that I tend to generate is synthetically originated. Should we say so my guess is one of two things. One, the model is judging you and saying that is not you probably you do not look as good as that. Yeah. And then the other side of that, if there's any kind of like product branding in there, or if it's seeing it as. As a real person, or if any one of those supermodel dudes have like famousness or anything like that, you're probably getting some pushback on that part of it. I've run into a couple of things with 2.5 where like 1 of my tricks that I'll do is if I'm running in an omni reference mode, for example, and I need like another shot from within that, I'll. I'll screenshot literally something from within there and then move into image to video. And I've had a couple of weird instances where when I've done that, I've gotten kickback on something that it's generated. So I think that it sees whatever it was trained on. Possibly.
- [05:54]
A
That's interesting.
- [05:55]
B
Oh, that's not. Wait, that's not. You're not allowed to use that. We can use that. You can't use that like so.
- [06:00]
A
Well, let's. Let's dig in a little bit because you made a whole short film and I do want to hear a little bit specifically about kind of what you've learned about prompting the system. One thing I will say with Sea Dance two that I found is that like there is a world where you can kind of give it the video that it generated and use that as a reference. And sometimes that works really well and other times it makes it really screwy. But what tell us first about the short film you made. Because it was great. We'll show some of it here. But also like what. What have your learnings from prompting the system so far?
- [06:26]
B
Yeah, so well, the film was Death Walks into a Bar. It was a. It's like a two minute, ish neo noir kind thing with a twist. Shyamalan, eat your heart out. And yeah, it's a very Twilight Zone y. I think that. I think that many members of the AI for humans viewership will enjoy it. So that was generated entirely with Sea Dance 2.5. There's a couple of things with it. First of all, that long 30 second output is expensive.
- [06:56]
A
Oh, we should talk about that. By the way, it's 1100 credits in my experience, which is like I. If you just like I. Part of the Dreaming CCP Creative Partners program. I think I get like 15 to 20,000 credits a month. As part of that. But that's. Yeah, that's what, 15 to 20 outputs, which is freaking crazy.
- [07:15]
B
I do expect that to go down and I do expect there to be crazy sales left and right. So I've seen some rumors and some rumblings that as the API goes out fairly shortly, the API pricing is going to be a little more on the aggressively saving side. I think by dance is. Is aware that Mini Max is out there. Like, like we have, we are now in a competitive marketplace which is great. So yeah. So I expect the price to go down. The 32nd thing I think is really great for sort of like, you know, my film is very two guys talking at a bar. So it's really great to do these long chunks of like dialogue back and forth because it's really good at, at that kind of thing and not doing like necessarily like having a big action sequence in the middle of it. I didn't really test that too much, but actually no I did. And it does get a little bit on the, the wonky side. See I, I do watch the show. Walkie. Walkie.
- [08:15]
A
Pretty good. We haven't seen him forever. Yeah, exactly.
- [08:20]
B
I do have a wonky. I have a good wonky walrus output for you later on. So. Yeah, no, I think it's, I think the, the, the cost is a big issue there and but I will say that I was able to. That entire short was done in about 15 generations which is pretty low. So that is. So there's a weird balance of like it may cost more but it's gonna, it's gonna cost you less as well. Yeah, possibly. Or save you more time.
- [08:50]
A
So I have this experience actually. I saw a really great video. I'm sure you're familiar with Dr. Whatever his name is. Bambi the Destroyer. Yeah, yeah. He just put out a video today where it kind of shows all the edits he makes in the mid. In the videos he makes. And I think that's one thing a lot of people who don't make these videos don't understand is that like the best versions are not just like stringing together four outputs and like hey, here's my video. It's really a lot of care and time that's put into timing the edits, making sure you have the right outputs. One thing that's interesting about 30 second outputs to me is if they're consistent it does give you more shots to play pull from, which is an interesting idea that. Was that your experience?
- [09:26]
B
It was, yeah. And actually if anybody like we can link to. I did A whole behind the scenes breakdown of that film on. On the channel. Yeah. What I had noticed is that, like, there was a few where our. Our book. Our character that plays death. Spoilers to those who haven't seen it.
- [09:41]
A
It's called Death locks into a bar.
- [09:43]
B
That's a good point. That's a good point. It's in the title. The. The twist is in the title. Um, the, um. That's eat your heart out, Shyamalan.
- [09:51]
A
Yeah, you should have just called it the Ghost. The kid is a ghost. The kid is a ghost. Bruce Willis is a ghost.
- [09:57]
B
Yeah.
- [09:58]
A
Sorry. Sorry, everybody. I spoiled something from 40 years ago. Now we're 30.
- [10:02]
B
The village is in present day. That should have been the title of the movie. Would have saved us a lot of time. So that. Yeah, I mean, there was a couple of outputs where the book was just like comically oversized. Yeah, I saw that one, actually.
- [10:12]
A
That was very funny to me. Like, it's in that way. Yeah.
- [10:16]
B
But, you know, at the same time, that was 10 seconds of it, so I still had 20 seconds of play that I could kind of mess around with. On the prompting front, by the way, I will say that it is a very different beast than 2.0, I think. 2.0. You were able to be a little lazy and sloppy with your prompts and just like, guy sits down and does this and like speak very conversationally. That does not feel like the case, at least at present. With 2.5. 2.5, they released two kind of like big tones of prompting structure for. Oh, wow. Yeah. And like, just like, you know, docs basically, like, you know, prompting and I mean, they're the instruction manual. So about midway through that production, I actually had to read the instruction manual, which I never do. I'm the guy that throws the instruction manual.
- [11:00]
A
Oh, me too, Me too.
- [11:02]
B
And I was like, well, this isn't coming out good. What's going on here? And yeah, so reading the docs, incredibly helpful. It feels like with 2.0, I often said that, like seed, which is their LLM, they snappy, fast.
- [11:16]
A
Really good.
- [11:17]
B
Yeah, very good.
- [11:19]
A
That.
- [11:19]
B
I. I don't know if we've ever fully gotten confirmation on it, but I think that that is baked into 2.0.
- [11:25]
A
Oh, interesting.
- [11:26]
B
I do not think that. I don't want to say it's not built in, but there's something. There's like the wires across there. It feels like a little bit because you have to be a lot more. It feels a little more like old school. Like old school. Like Runway three Style Prompt.
- [11:42]
A
Interesting. Okay. Yeah. I think one thing I found really interesting is, and I'm sure you're doing this in some form too, like I am using rather than prompting it raw. What I'm often doing now is using a Claude system that JBoogs Creative has created in some ways like using things called YAML prompting or JSON prompting. And that does give you often better results. Well, that's good to know. This is. So it just has come out today. It's now available in America. We should move on to the next other biggest thing, which actually I think might have a bigger impact, even though it's a slightly worse model, which is Haluo, which is the company that has made the. What else does Halio make? They make.
- [12:25]
B
They've done. I mean they do everything. They have a. They have an LLM, they have an audio model, they have a couple of agentic things.
- [12:33]
A
Yeah, that's right. Yeah, yeah, they've done a fair amount. So, yeah. So this is Minimax, right? So we've been talking about the Minimax model for a while and Minimax is. This new version is called H3. It is very good as well. It is a little closer to what you would think of as Sea Dance 2. It's not 30 second results. I think it's like. Is it.
- [12:50]
B
Oh, yeah.
- [12:51]
A
Oh, okay. Well, hold on one second then we'll go get that in a second. The most interesting thing I think about this, Tim, is my experience with this is it was pretty good. It felt a little bit like Sea Dance too fast to me when I was using it, which was when I was using it just for regular stuff. But. And I have not done this yet. I have a PC that's in a box that I have to open up. The biggest thing about this, you can install this and run it locally on what seems like pretty decent hardware. Meaning that like you don't have to have like a server system to run this. So let's talk about this. You have done this locally, is that right?
- [13:26]
B
So I actually do need to do a quick tutorial on that before I go on vacation. I think so, yeah. It is very easy to install. I know that a lot of people are like, you know, they have Comfy anxiety. Anytime that, you know, somebody sees Comfy, they're just like, I don't know. This is. Is completely insane. Comfy has done a lot of really good work in kind of.
- [13:44]
A
Oh, we should just define that for anybody who's new to this AI space, which sometimes every once in a while on our show There is Comfy is Comfy ui, which is kind of image
- [13:51]
B
of the most confusing node system ever.
- [13:54]
A
Yeah, they call it. It's. It's the spaghetti nodes. If you've seen a bunch of people moving boxes around with lines into things, that's Comfy. In a lot of ways it might be something else, but mostly Comfy is a system that is set up so that you can very specifically send prompts and things through different models and different stuff. And it basically allows you to create a kind of an automation system for specific models. Right?
- [14:16]
B
Yeah, I think that there is a little bit. Like, the guys that are really good at Comfy are. It's like they're unicorns. Because I think in half, like it's half of a creative brain and half of an engineer's brain, like, it could be really, really good. Luckily, we now have Claude and MCP and CLI tools, so you can actually use your cloud bot to go over there and say, like, get in there and make me the most awesome node system of all time. And it'll do it. But Comfy itself has actually made the installation process and all of that because previous to this you had to get GitHubs and you were chasing around doing terminal commands for.
- [14:55]
A
Oh, by the way, it's a night and day, right? My experience has been the last three months, I mean really since Claude code kind of came on at the end of last year. But I have installed Blender and had it work with Blender and I'm not a technical expert and I've been using Blender but. But really for open source models. And Comfy, it is like, it's not plug and play. You'll still have a couple things you have to fix, but the system can do it. Right? Like, that's the amazing thing about it.
- [15:19]
B
Absolutely. So, and then Comfy itself has set up templates that are one click install. Once you're there, you just, you hit H3 text to video, reference to video, image to video, boom, you're up and done and running. And Yeah, I mean it's been, it's been reported. I haven't actually seen it, but to output on like a 3090, which is insane. That's.
- [15:41]
A
Which is crazy. So, yeah, if you're not familiar, that's like a graphics card that's a couple years old, right? Yeah, my graphics card. I have a. I have one of
- [15:49]
B
these Hyundai Elantra graphics cards.
- [15:51]
A
That's right.
- [15:52]
B
Dependable, it'll work.
- [15:54]
A
But 20, 21 maybe. Yeah, I have a pretty beefy laptop PC but it has like a really high end laptop graphic card. So I'm, I'm gonna pull this open one question we're gonna get into like what you can do with this since it's open and local, which is a big deal. And that's what you saw on our keys. But yeah, you could do a lot.
- [16:14]
B
You do a lot.
- [16:15]
A
Tell me what hardware are you running it on and like what. How long does it take to put out to, to create one? Because people might be surprised. Like you think you have to wait a while for an output from C Dance or from other places. If you do local video it takes a little while longer to get the results.
- [16:30]
B
It does, yeah. So I am fortunate enough to have a 50 90. I generally, if you watch the channel I'm on my, like they have the Mac here and then it's, that's plugged into the PC over here. But I would say like generally a smart idea is to generate at 720. So you don't want to be trying to do 4k or even 1080. Really you can as doc. As in the documents, you can generate up to 15 seconds. Okay. And I would say at like 7:20 if you're in around like 10 seconds or so, you're probably looking at like five to eight minutes or so. So it's a little bit slower but there are lots of like little adjustments that you can make in there as well where you can like you had mentioned that it kind of looked a little bit on the Sea Dance 2.0 fast side. If you're in Comfy and you ask your Claude bot like hey, how do we bump the res. The quality of this? You can, you can play with numbers like steps and whatnot in there. So it'll. Oh yeah, longer but it'll also come out with like kind of more of a, a stable video. It's all the sliders.
- [17:36]
A
Now I haven't thought about step stim since probably late 2024 when I was doing stable diffusion 1.5 stuff and I was trying to figure out how to make a bunch of like old men video old men images of myself.
- [17:49]
B
But anyway, so real quick. Interesting. Yeah interesting point in there when you're in Comfy, you do have all of these additional controls in there as well. I have found that you can break that 15 second limit. Oh wow. So I have been getting it to generate. Granted not a lot happens but I just out of curiosity I was like let's, let's crank this up to 45 seconds and see what Happens. And it took about an hour, but I would get videos of like my FBI agent in a diner, like at 45 seconds. Wow. Common test that I run again, it took about an hour. He kind of goes through, through the allotted time a little bit over maybe about like 25, 20. About 20, 25 seconds. And then they just kind of sit and kind of hang out. Like they're not really doing a lot.
- [18:32]
A
It's like, why are you asking us to do this? We're here, we're here.
- [18:35]
B
Like, we have no more lines.
- [18:38]
A
So.
- [18:39]
B
But it is, if you need room tone, for example, in your AI short, it's a good way to do it. So it can be done. So. Yeah. Yeah.
- [18:46]
A
Well, let's. Okay, so let's talk about what it means. First of all, local means two things to me. One, it means you are not limited by price, you're only limited by time. Right? And one of the things that if it's going to take five to 10 minutes or longer for each output, you can queue a bunch of them up, go to bed and wake up with a bunch of videos, which is always a nice kind of thing in the best case scenario. The other thing it means, and I think this is the thing that I think is more interesting and kind of. I tweeted about this yesterday and it does make me think like we have kind of crossed the Rubicon, basically. It is uncensored. And what I mean by that is if you go to the Haylou minimax website, which, you know, you can go there and you can generate stuff there you are being filtered by their bots and by whatever their front end says here we'll show a few of these. But there are amazing outputs people have had of things that are in the training data. Things like Friends, things like Seinfeld, things like Family Guy, things like the Office. And it is not perfect, but it is kind of like when we first had that Sora to moment that came on where it was like, you know, everybody's able to do anything with anything. And because it's maybe not as mainstream as ChatGPT and it's not an American company, it hasn't been as big a deal. And by the way, it's not as good. Right. I think there are moments even in The Sea Dance 2, you know, where you have the Tom Cruise, Brad Pitt fight. They put the restrictions on it. What's interesting about local models means that you can download the weights, you can have the whole thing on your computer. They cannot restrict it once it's on your computer. So I want to talk a little bit about this and kind of learnings about what you've seen from this side of it.
- [20:22]
B
Look, yeah, I mean, it can be done. And we have seen, like, everybody loves to generate an office clip. Like, we've seen a lot of those and we've seen a lot of like the Family Guy clips and your Darth Vader's and your Stormtroopers. You know, to me it's like, look, I guess you can do it. I don't know how interesting it is. Like, you can do.
- [20:41]
A
This is a good point.
- [20:42]
B
Yeah, I love this point. Anything that you want, like, you can make anything that you want. And like your go to is like, Stewie, like, really? You know, I mean, like, why?
- [20:52]
A
Well, okay, this is a really interesting point for the AI video world at large because later on we're going to talk about some of my favorite AI creators who have come out with some really interesting stuff in AI. See what you did there? But, like, I do think there was this really interesting moment in video and AI right now where there's people that are kind of rushing to get the video out that will kind of get the clicks right. And really it's, I think, mostly on X, but other places as well. And I think that there is this burgeoning world of people that are learning how to use these tools, but are learning how to use them mostly for attention and not necessarily for filmmaking. And I think that I'm trying to, in my kind of side projects I'm working on, try to get better at understanding. Like, okay, how can I tell a story that I might want to tell that I don't have the money to go out and like, create a whole film for? But, you know, it's an interesting balance, right? Because when you see. Maybe we just listened to this clip from Seinfeld here. Will, why don't you put this sound up from the Seinfeld prompt?
- [21:47]
B
What? Now? Here's Johnny Kramer. What are you doing? I'm home.
- [21:57]
A
When you see something like that or you see Walter White meeting Joey, there is this kind of like shock to your system because you have this kind of collective understanding of who these characters are.
- [22:08]
B
No, and that's, you know, that is clearly what. And I mean, no diss against who, whoever made the Seinfeld clip. I mean, there's. There is. There's a couple of different sides to that. Like if it's a clip that you're making of Seinfeld's or whatever, and you're sending it out to your friends Sure, I get it. You know what I mean? Like, it's like, it's fun. And then there's the other side that is sort of like I am posting these things to elicit a reaction to get.
- [22:34]
A
Yes.
- [22:34]
B
Hits on likes to get the. Likes to get the. To get the sweet, sweet monetization juice of any one of these platforms. And it is sort of like, I guess there's like a side business of sorts to a lot of people where, you know, you're starting to edge into the content farm slop factory. It just looks a lot prettier and you're utilizing and you're sort of piggybacking off of established properties and I don't know, I just don't think that's, I don't think that longevity wise, I just don't think that that is going to. That's a good play.
- [23:06]
A
Well, and also there's interesting versions of this. Like, yeah, I'm sure you're familiar with Charles Curran who does use like, you know, real images of people or different things, but they're like almost like commentaries on like what's going on. And he's doing like kind of unique things. He's not just recreating an office environment. One thing I will say that I find funny about these, he did the
- [23:24]
B
Goompocalypse today, I think.
- [23:25]
A
Yes, yes, the Goompocalypse, which is very funny, basically. I mean, he's again like Jay Bugak's boo's creative, like kind of on the edgier side, but still really interesting and good at these tools. I do think it's funny. Before we move on to the next story, the gym from these clips has the weirdest eyebrows. If you look at them like, it's like every eyebrow is a little funky and that's just kind of where we are. But, you know, okay, let's talk about WAN3 and then just like kind of what they've announced because I know you haven't used it yet, but Won three is another Chinese model. And just so you know, Sea Dance and Minimax are Both Chinese models. Won 3. What's cool about this? That just kind of got announced and came out today.
- [24:03]
B
So this is not going to be local, I don't believe. At least I think they're touting this as a 60B model. So you, you aren't going to be cramming this onto your home machine. But it's got some interest. I mean, like, Wan's always like Wan 2.2 has always been the darling of or has long Been the darling of the open source community. Like so 2.2 was open sourced and a lot of really great stuff was built off of it. But itself as a base model WAN has always been very good, very competitive. The interesting thing that they have in this one that I'm very curious to test out is that apparently it can take like text documents, which is pretty wild.
- [24:45]
A
I'm so excited for a Terms of Service video which I talk about.
- [24:50]
B
You can do. You can put a PowerPoint in there, you could take an MD file in there, you could put an XLS file in there. So you can put a spreadsheet in there. PowerPoint quarter.
- [25:00]
A
Quarterly reports are getting very different. Yeah, get ready.
- [25:03]
B
It's good.
- [25:04]
A
And quarter. It's good.
- [25:07]
B
Which I know it sounds like that's the most exciting thing in the world immediately. Like there are some. This is my deep cut joke. There are some EVE Online players that are really interested in AI video all of a sudden because that's the spreadsheet simulator.
- [25:20]
A
Of course. I love EVE Online. I've never played it. I've just been fascinated by it. I'd like to try. I would like at some point somebody to take a weekend and just walk me through it. But I would have to get an extra weekend in my life so that my family wouldn't look at me and be like, what the hell are you doing?
- [25:37]
B
That is a game that I do not play as well, but I watch endless documentaries on YouTube about, oh, it's amazing.
- [25:42]
A
The economy of EVE Online or so if you're an EVE Online fan, drop it in the comments and tell me. You might convince me to try it this weekend.
- [25:50]
B
So, so, yeah, but I think that going back to what you were talking about with, with J. Boog and his JSON prompting and Yamal prompting, I think that having, weirdly enough, I think like having spreadsheets in there might open up doors to very interesting different ways to instruct it out. Yeah, yeah. So I think that'll be. I'm curious, I'm curious. I don't know where it's going to lead. I just think that they just, they, they, they, they, they created a door and no one knows where it's going to go.
- [26:19]
A
So I mean it's really interesting. Okay, we, we're probably not going to talk a lot about Flux 3, but Flux 3 is out. I, I've played around with it. It's very cool. One of the coolest things I love to do with it is 80s and 90s videos. Purrs, Beats, who is one of our favorite AI video creators, comfy Master, if you talk about Comfy, has done some really cool stuff. Do you have like a line that you want to say about Flux 3 at all? It's available on foul. You can use it. You can. Anybody can go try it right now.
- [26:43]
B
Yeah, it's. It's good. It's been long overdue. We have been waiting for it for quite some time. I just need. It was announced actually with. When Flux one came out. It was like they had their little.
- [26:56]
A
Yeah, I remember.
- [26:57]
B
Yeah.
- [26:57]
A
Yeah.
- [26:58]
B
So it's been. It's been long stewing. Yeah. I mean, look there. I played around with it a bit. I did not kind of got a handle on it and it felt very again, kind of in that like mini maxi area of around Sea Dancy. And then I was in early access with it as well. So I was generating within their discord. I sort of did my first initial impressions with it and then everything kind of blew up. So I've got to circle back on it because I know that they were. They were. And they are still making updates on it. It. So.
- [27:28]
A
Yeah. And supposedly it's going to come out. There is. They have said that it's going to be open models as well. Right. So I guess we'll see when that drops too. Okay. We should talk very quickly about what it means that. Tim, except for Flux 3, the three things that we just talked about are all Chinese models. And I do think it's an important conversation to think about. And. And I'm just curious to hear your kind of first thoughts. Does. Does it matter to you if a model in AI video is Chinese versus American?
- [27:53]
B
Does that.
- [27:53]
A
Do you care at all? I just. As a top line.
- [27:56]
B
At the top line, I don't. I'm going to use the best product that provides the best results. Like there's no. As we. As we. As we said in the 90s, information knows no. Knows no borders. So, you know. No, I'm. You know, I'm gonna jack into the old cyber. I'm gonna put my. My cyber suit on and jack into the net.
- [28:18]
A
That's right. What other else could we reference from the 90s? Let's see. But we got friends. We got other things.
- [28:24]
B
We got friends. Yeah, that's true.
- [28:25]
A
Well, that's makes sense. So my next question would be. Is the Lawnmower Man?
- [28:30]
B
We need to talk about the Lawnmower Man.
- [28:32]
A
Yeah, exactly. We're gonna. Yeah, put us in lawnmower man. Like 3D models. We are gonna talk briefly later about how Google Seems to be kind of struggling and that Demis Hasabis has been moved off and Jeff Dean, one of the founders of Google, has left. Are you worried at all about, like. I guess worry is the wrong way. If, if you don't care if there's American versus Chinese models in this way or you're gonna use whatever. Does it feel like Google or any American company is going to pursue this or do you feel like. In my mind, I kind of have a feeling that, like, most people might be seeing like, AI Video as just too much of a hassle. And I don't mean that like, listen, I am a WGA member and a union member, and I understand all the issues that have come up around AI Video and the SAG actors especially understanding that, but does it feel like Americans might just be like, you know what? This is not worth it for us.
- [29:21]
B
I think that there are, there are law firms that are really hoping that all of them decide to. Because they billable hour. No, I do think that there is with Google. I'm curious. You know what I'm curious about is, Is Meta, because I think that, yeah, Meta has. Well, they paid all them bucks to. What's. To our boy, what's his name?
- [29:45]
A
Alexander Wang.
- [29:46]
B
Yeah, yeah, yeah. So they've got a. So, you know, he's gotta, he's gotta make a splash at some point or other. I, I know that they're floating around with that. They released Muse not too long ago, which is pretty good. And I know that they're, they're. They're. I mean, Zuck's always working on it in the background. Like they're.
- [30:04]
A
Yeah, it's interesting. I think Muse. I tried their. I think it was their Muse. Spark is one of their models. I remember that's the name of their video model. Their video model has not come out yet, if I remember. But the image model is out. So the. And their image model is fine. What was interesting about Muse to me is it kind of felt like. Remember when Elon got serious about Grok and kind of Grok one was not that great, but by the time you got to Grok 3 or something, it was like kind of close. Right. I do think that Zuck and Meta is like deep in the back end of this now. Like they're going to spend the money on it. So it might be that we get a model, a video model from them. That's pretty good. What they showed in the blog post when it came out, felt like it was a couple of generations behind. But you can catch up pretty quickly
- [30:44]
B
very much like, is a big deal. Well, I think Elon's a really good example of that because Grok Video, Grock, imagine is they just released their 1.5. And like, if you've been. I haven't really covered it on the channel recently because every time that I mention Grock, like, there's a. There's a contingent of people that, that are absolutely, you know, anti.
- [31:03]
A
Anti. Elon. Yes, we've been there.
- [31:05]
B
And which, by the way, I might be.
- [31:08]
A
I'm not saying there's definitely a lot of things that have gone wrong there.
- [31:12]
B
Yeah, I'm not, Look, I'm not. I'm not like the. I'm not the most Elon. You know, I don't, I don't march to the, the beat that Elon lays down all the time. But he is sort of. I mean, any of them really. I'm like, that's a whole other topic. But I mean, like, each of them have their own. Their have problems. But yeah, the, the. As quickly as Grok ended up just leveling up, like, I mean, the 1.5 model is. Is pretty good. So I definitely think that. And a lot of that actually, I think secretly has to do with the fact that you remember when he found the vine archives? He found.
- [31:47]
A
Oh, yeah, sure. Yeah, yeah, yeah.
- [31:48]
B
Just tons of video to train on. And like, who else has tons of video to train on? This is a good old. Good old Zuck.
- [31:55]
A
Yeah, Zuck has a lot of it. And by the way, YouTube as well. Why you always can't leave out Google. But also the creator economy is really interesting. I don't know if you've caught that Hank Green conversation that's been going on, which is really, I think, a bummer. And if you're not caught up on this, the very quick TLDR is that Hank Green basically, in an interview with somebody, used a phrase that sounded like. It's like, I give. I'm going to push back on you. Which is a AI like phrase. He did not use AI to write a script. He did not use AI to do it. He just literally used. He's been using AI to help him research stuff and like kind of maybe ingested that phrasing into his brain. Anyway, the creator economy, I think there is a very anti AI kind of cohort. Now, Google as a whole, I think is pretty pro AI video. But if you've been following this, they've also canceled a bunch of AI first channels. Now, a lot of those are very sloppy in some ways, and I Think it is a bad thing. But anyway, we'll keep an eye on this. Like, I really do think if Google doesn't come out with a Omni, full blown Omni model, which is, you know, the Omni Flash came out, but we do not have like the Omni Pro model. There may be something, a pause on that. But before we do that, I want to make sure that we get everybody. We're going to take a promo break here, Tim. Promo break, Promo break. We're going to, we're gonna promo. Go follow Tim at Theoretically Media. And you know what, if you're here and you're Tim's fan, you just click our own subscribe button. Click, click the buttons down here. We do this every week. We, you know, we're in a mix of like doing it once or twice a week right now because it's the summer, we're doing about once a week. But in general, we do this every week. Like Tim, we are independent. We do this on our own. Kevin and I both have been putting a bunch of our own money into this. We appreciate it. Go join the Patreon if you want to throw us a couple extra bucks. But yes, click, click, subscribe both here and there. Wherever Tim is over there. Okay, Tim, I do want to. Let's, let's move on here. I want to talk about something else in AI video. That is a big deal to me. What I was happy with is this got a lot of views on your channel. You had a really good video about Higsfield. And by the way, I also want to shout out excellent thumbnail. You are getting very good at your thumbnails. The yellow with the word ugg at the top. Tell people what. First of all, what is Higgs Field? We've talked a little bit on the show, but what is Higgs Field and what was your video about? Because I have a very specific feelings about Higgs Field that align with yours. But I do think it's a really interesting thing to think about this, especially when it comes to where you are going to get served these AI video models.
- [34:19]
B
So Higsfield is an aggregator platform. So they take the APIs of your sea dances and your nano bananas and your clings and everything. They kind of wrap it up into. And look to their credit, I think that the, the platform UI itself is. Is. I have problems with it, but it is well designed as well.
- [34:45]
A
Oh yeah, it's a good, it's not a terrible way to use AI video.
- [34:48]
B
I will say that that's been my experience, but it is also built very much like a Vegas casino. Know, in that when you walk in the door, it's. There's stuff and there's. Oh, it's all. You're overwhelmed. And then, you know, 700 later you're like, where's the exit? And you can't find it. And again, that is all. That is all 100 by design. On top of that, they, I gotta, you know, I don't. Again, I, I don't want to be Higgs Field bashing guy. I did the one video that was essentially they, they had. They. There's. They have a fairly documented history. Alleged documented history that you can look up over@higsfieldsucks.com I am not affiliated with that site. That is just. That is. That is. Somebody set that up. And then additionally there's a. There's a kind of Gloves on Forbes article that was written about them. So if you need to get caught up on all the Higsfield drama, feel free to. The, the. The thing that I kind of took issue with, I guess is that. And kind of went like, kind of became wildfire on the Internet was they released updated terms of service and somehow or another I, I'm on their email list. So I ended up grabbing it and just like, just knowing them, I was like, well, what are they?
- [36:03]
A
Let's see what's in here. Yeah. So.
- [36:05]
B
So I sent it over to my, my law firm of Claude Fable and Ultra Code. Very good lawyers. Very good lawyers. So. And they read through it and they had like, I mean like there was like Fable did not like things about it and, and gave me a bunch of notes on it. So. So I released those out to the public and it kind of created a storm.
- [36:26]
A
Yeah.
- [36:26]
B
So yeah, there is.
- [36:28]
A
I think it's really interesting. So, so just the TLDR here is that basically I think there's a lot of talk in this terms of service about like Higgs Field not only kind of like using the prompts that you're using to kind of train. Right. But then also this idea that like they could train on your videos. But also there is some ownership questions too. Now again, you know, the ownership of AI video is a big question mark. I think in general, just to remind everybody there was a copyright case that a lot of people who hate AI cited that like AI video cannot be copyrighted. But that was because it was only. There's an exact output of an AI. You can still create a character, trademark a character. Create a character. Like our friend PJ Ace actually knows a lot about this stuff. What specifically? Like, maybe one thing in this stood out to you that was kind of crazy and made you kind of like,
- [37:19]
B
there's a few of them. But the. So nothing on. Well, I think the two big issues were that a lot of creators had was that if you were to generate. And I do want to be clear, they did revise their terms.
- [37:33]
A
Yeah, sorry. They did revise it, which is good. And they listened, which is smart.
- [37:36]
B
So, yeah, there is, there was. Yeah, the happy ending. Ending to this is that we all pushed back and. And, you know, their hand was in the cookie jar. We slapped it and they, they. They pulled back so. Good. In the initial terms of service, there was two. I mean, I think two of the bigger problematic things that were in there were that if you generated anything on their site, they essentially had an open license to use it as promo for anything, which I. I know. I don't think that's a, you know.
- [38:00]
A
No, exactly.
- [38:01]
B
Yeah. Pay a license fee.
- [38:03]
A
Yeah.
- [38:04]
B
And the other thing was, like, biometric data kind of came up, and that's always super scary. It sounds like, you know, very blowfield when you're just like, I will have all of the biometrics. And it really wasn't that. But they. But snuck in. There was like, if you take a photo of yourself and put yourself at the Lakers game, which is, I think, a template that they have, or, like at the Yankees game or whatever it is that they don't own the likeness to. Like, they don't own your likeness. They are.
- [38:32]
A
They have the ability to use it.
- [38:35]
B
Use it. Yeah, yeah, yeah.
- [38:37]
A
So anyway, I mean, I think it's. It's an interesting thing. And yes, as you mentioned, they did change it. I do think we're going to run into a lot of things like this going forward.
- [38:45]
B
Two things with that. One, I am not terms of service guy. Like, people, everyone's like, like, what do you think?
- [38:50]
A
Okay, go over the chat.
- [38:51]
B
I don't, like, I'm like, I'm done. As I said, my piece.
- [38:54]
A
I'm done.
- [38:55]
B
I know, right? And like, look, there are problems. I will say this blanketly. There are problems with terms of service with all of these platforms. And yes, of course, I think to a large degree, we're at this point. We are at this point. That kind of reminds me a little bit of the terms of service war that we lost with social media. We all went, you know, when Facebook first came on, we're just like, yep, I will. I will take that. Like, free toaster. Yes, sir. Here's here's everything quick? Yes. And we all lost that fight. I hope that we do not repeat the same mistake again. So this is a good moment for everyone to collectively say, like, what's in that tos? At the same time, I am not the one that's going to be out there, like, running up that hill because I've got. Yeah, no, this is not my job.
- [39:41]
A
You know what I'm wondering? This makes me think because I'm in the market for something right now. And by the way, I don't mean. I don't mean this to be an ad. We've not taken any money. And if you have, that's okay too. But, like, where are you doing most your generation? Like, what I have found is I'm kind of spreading myself out across a bunch of different platforms. But like. Like, is there one platform that you really like?
- [39:59]
B
I mean, there's a couple. Like, I circle back and forth just because of nature of the job. Like, I'm sort of all over the place, so.
- [40:05]
A
Yeah, me too.
- [40:06]
B
Yeah. I mean, like, what I call the motherships are always good. Like the. The main platform.
- [40:11]
A
Yes. Like mini maxes or. Or dramina for.
- [40:15]
B
For.
- [40:15]
A
For Sea Dance.
- [40:16]
B
Right. Yeah. Yeah. For the aggregators. There are. Yeah, there are ones that, I mean, I think that all of them are pretty great. Magnific is always great. There's sort of like the small. I call them smaller tier ones. Open Art is really great. And Flora, I really enjoy. If you like stuff. Flora is really a lot of fun. There is another one called Artcraft that I really like. And then I guess closing out would be Martini. I actually really like that a lot. That is. Oh, that's cool. Yeah. Martini is one that's set up by.
- [40:46]
A
I've heard of that. Oh, that's right. This is the one that's by. What's his name?
- [40:49]
B
Right, the guy.
- [40:50]
A
Yeah, yeah.
- [40:51]
B
Who is an. Who is. I shouldn't say X.
- [40:53]
A
He.
- [40:53]
B
He is a dp, is a director of photography. So everything in it is very much built as it would be through the lens of someone that is. That is a camera person. So, like, you can be that.
- [41:07]
A
That's the subtitle for our episode.
- [41:10]
B
Yeah. Just, you know, you can. You can definitely feel it in the platform that it comes from a filmmaker standpoint. So.
- [41:15]
A
Okay, well, we've talked about for a very long time. Which just makes sense because you're here. I want to hit two very big stories quickly that are kind of more general AI news. First of all, we are very excited. The rumors that are coming through synth waved, which is a funny thing that there are certain ex people that are more reliable than others. But I do want to shout out, this guy has been very reliable. He is saying that we are getting Astra, conceivably next week, that they are targeting it, which is the new OpenAI model. This is the one that Sam Altman has brought to Washington supposedly to show off. This is also the one that is rumored to be better than Fable. The interesting thing about this tweet from him is that it is a new base model, which we have been assuming, but he said it is the largest since 4.5. And if you are an AI aficionado, historian, whatever, you remember that 4.5 was one of our big kind of moments of like wah wah in the AI space because it was a huge model train, but was kind of disappointing and very slow. Tim, I guess this is a question more about. Maybe it's more of kind of the OpenAI cloud question for you. Like, my experience has been so far that I am kind of jumping between the two of them. They are clearly right now the frontier. Right. Yeah. And I've now developed this. I don't want to call it a relationship because like my Claude code and I have become like buddies. I've been making these weird videos with my Claude code. I named him Fig. And oh, there's a new website for Fig and Moss. If you're go to figandmoss tv, you can watch all of my Fig and Moss videos. I don't really have that with OpenAI, but I'm wondering if you have a preference or are you kind of a bounce back and forth kind of guy too?
- [42:45]
B
I am a bounce back and forth. I tend to use. I tend to be in quad code a lot.
- [42:51]
A
Yeah.
- [42:51]
B
Because that's sort of where it's like I got my fingers like. But I have that cli'd over to Codex as well. And then I'll sort of bounce back and forth between the two. Lately. I know that there's been a lot of like Opus pushback, Opus 5 pushback lately. Like, it's not. It's a lot of people are like, it's. It's a little. It's a little dumb. We don't really like it. I will say from my standpoint, I. From a personality standpoint, because they all do have certain personalities. So I do find Opus 5 to be less annoying than Opus 4.8. So, you know, just to sort of like chat with or brainstorm ideas and whatnot. Generally I'm Running Fable Low.
- [43:32]
A
Oh, interesting.
- [43:33]
B
Okay. Yeah. So my corner, my. My mundane tasks. I.
- [43:38]
A
You know, it's so funny. I've never used Fable Low, but do you find that it stretches the credits quite a bit further for some reason? I am always on high. But do you find it's great in that way? Like, does it actually give you a lot more?
- [43:49]
B
I do. And actually the two things that I have hammered into it just over and over and over again are you are not allowed to burn. Yeah. You were not allowed to burn Fable tokens. That is gold. You use your stupid sub agents of haiku and. And you use on it or whatever. Yeah, yeah. Use them as much as possible. And then the other thing that I've. I've told is like. And just hammering it into it over and over again is like, when we are on a task, you are working at low as often as possible.
- [44:18]
A
Yes.
- [44:18]
B
Until you know that it's time to kick up. And like, it basically has never told me, like, this is an Ultra code test. Once or twice it has. Like, Right. And even then we learned that lesson because it burned through a week of Ultra Birth through a week of Fable. When, like before we had Fable.
- [44:34]
A
Yeah, yeah. There's a nightmare when you're like, oh,
- [44:37]
B
no, I gave it a task and then it decided to sub agent it. Sub agent a bunch of Fable agents out. I'm just like, you know, but interestingly, that was the week that that soul came out. So it was a good way of like testing both.
- [44:54]
A
Cool.
- [44:54]
B
So yeah, soul, I. I like as well. I think that GPT is or open AI, especially if you get into the how you talk to me part of it and how you work with me. It. It works really. Like, it swears at me all the time. It. It basically talks to like an, like an old detective to me, which is great. Like an old dumb.
- [45:13]
A
I love that. Which is fun. Yeah.
- [45:14]
B
So. So, yeah.
- [45:15]
A
Well, cool. That's awesome. Okay, very quickly, we. This. We touched on this briefly before, but there was big Google news that Demis Hasabis is stepping down as CEO of DeepMind. He's moving up to chairman, which. But he is going to be focusing on the future of AI. And Jeff Dean, who has been at Google for 27 years, is leaving to start his own kind of recursive intelligence company. There's a lot of rumors out there going around that this might be Google signaling that like, hey, we're not going to play the frontier anymore. We've been kind of watching that. I saw a note just recently Today from. I can't remember the guy who was. Oh, Alex. Anyway, Alex, who used to be at the Verge, now has his own newsletter that they are kind of disappointed maybe in Gemini 4 so far. So like, this might just be a signal that Google is going to say, like, look, we're not going to compete with OpenAI and Anthropic at the Frontier. We are going to say that we are a provider. We are this. I don't think that's a bad place for Google to be because like, they're the backbone of lots of things and maybe they're just like, you know what, what, we're not going to be that company. But I don't know, maybe in the long run that is a bad thing.
- [46:22]
B
I don't know if they need to though. That's, that's the thing with Google is that like, like again, AI is like a fraction of what Google is holistically. Like we have, they have gmail, they have YouTube, they have phones, they have, you know, I mean, it's like the Google, the, the, the, you know, the, the massive nation that is Google, like is, is, is much larger than just AI. So I don't know if they need to compete necessarily.
- [46:47]
A
I think you're probably, I think you're probably right. It'll be interesting to watch over the next like say three to six months. All right, now it is time for our favorite thing of the week. And this week it's an AI Video edition. It's AI See what you did there?
- [46:58]
B
Sometimes you're scrolling without a care, then suddenly you stop and sh.
- [47:15]
A
Okay, so we're going to talk through three videos that I've seen. Feel free if you know, I think you may know these guys as well too. But these are three really top shelf level AI video. Like in a lot of times, AI See what you do there. We point out, like things that make us laugh or goofy things. But these are three creators who I think are at the top of their game. And I'm going to start off with Kevin, the kid who is somebody that we have been following for a while the fourth episode of his video series, which is called Chronicles of Bone, which is a, which is magnific. Speaking of Magnific is helping bring to the forefront. I think they're probably paying him to do this work and work on this and giving him the tools. This video is what you should show your friends who think AI video can't be good. I, I, this is my take. Kevin has been doing this for a while. If you Remember way back when we talked about. I think it was like he did a fan video. Was it. It was a Batman,
- [48:07]
B
Star wars ones.
- [48:08]
A
Yeah. So stuff to me is just watching this video gives you a sense of just how big and crazy this world has gotten. And he is just. It feels like to me, this is a whole another level of what you can do with this stuff.
- [48:20]
B
For sure. I mean, I think that if you were to want. I mean, actually between Cavan Cavin, a very nice guy as well, I think that if you were to watch this and kind of have it on, you would not, you know, on your home tv because I think that this is something that you can do. You can watch this on your tv.
- [48:39]
A
It's designed for a big screen, basically.
- [48:40]
B
Yeah. There is, there is definitely not a lot that says I'm sorry, there's.
- [48:47]
A
Okay.
- [48:47]
B
There is definitely not a lot that says AI about it. Like you can almost. You can watch this as if you were watching a Netflix show.
- [48:54]
A
Yeah.
- [48:54]
B
I feel. Yeah, I feel that this is. We've definitely reached the point where what I like to call the. To be Tsunami. Yeah. When. When stuff like, when there's enough people like Cavin and Gossip and a number of others that essentially there is like, there, there will be the.
- [49:16]
A
There'll be a hit. There'll be a hit.
- [49:17]
B
There'll be a hit.
- [49:17]
A
There's going to be a hit at
- [49:18]
B
some point or another and we're going to see a lot of stuff come out that is at varying degrees of quality.
- [49:24]
A
You know what I mean?
- [49:25]
B
So. Yeah.
- [49:26]
A
Yeah. And I think that's another good example. Dave Clark, who we've talked about in the show a lot, Part of the Promise Studio released a vertical video with CDNs 2.5 that I thought was really interesting, this kind of found footage style. And one of the things that's really interesting about these AI video creators is that they will have a fair amount of success on a platform like X with 16:9 video or YouTube in Kevin's point. And our next creator we'll talk about. But there is this whole burgeoning world of people that are making stuff for vertical video too. Right. And I haven't seen Dave do a lot of vertical, but this was a really cool use case of found footage styles. Right. And being able to kind of open the door to that sort of thing as an AI video is cool as well.
- [50:05]
B
I really love the concept that Dave went with here too. Is that because it is like. I mean, my tent pole, like read immediately is just like, oh, it's It's Blair Witchy for the modern day.
- [50:15]
A
You know what I mean?
- [50:15]
B
Like, it's like, exactly. Dave knows what he's doing. Like, he's, he's making that call himself. But I think it's actually, I think that's really, it's a great idea. It's a great, it's a great format for AI video. Funny side story, by the way. A couple years ago my kids were like scrolling through Tick Tock and. And they came up to me and they were like, have you heard about this Blair Witch thing? I'm like, yes, I am aware witch.
- [50:37]
A
And like, like, it's all cycles, Tim. It's all cycles.
- [50:42]
B
They were like, it's really messed up that that happened to those people. And I'm just like that, that's a movie. So apparently at some point there was an account that was just repurposing the Blair Witch and presenting it to the Tik Tok generation as a documentary. So it does, it really does come all the way, like full circle. It really is.
- [51:00]
A
That's amazing. And finally, what I think was my first favorite AI creator who, you know, uses ip, let's say this is a Demon Flying Fox who made the Balenciaga Harry Potter videos, which really a lot of ways was like the first big AI video hit other than other than the Pope in the puffy jacket picture.
- [51:18]
B
Yeah, yeah, yeah.
- [51:19]
A
Demon Flying Fox has now kind of started to kind of like expand out a little bit more. He's making videos with original AI songs. But this video, he's so good about understanding like where the culture is and like what will hit this is basically the cast of the Odyssey. He's using real people. So again, there's a lot of kind of like funky IP stuff going here, but he is turning them into the Apocalypse now scene where they are going down the river. And it's just like such a cool look. It's such a cool thing to do with AI video. I think this is in my experience, unlike those clips we were talking about before, like the Seinfeld kind of quick, quick hits. This feels like like a piece of art in a weird way. I know people in our audience might be like, like, ah, it can't be art, it's AI video. But like this feels like it levels it up by using these things to me. And I just, I think he's one of the best people out there right
- [52:04]
B
now that gets into a level of like, what? I agree with you completely there, that there's a. There's. I think the difference is is like, if you take this into like hip hop sampling and DJs, like there's a difference between like, you know, taking a beat and then making something with it, which he did here, and then just, just playing the record. You know what I mean?
- [52:25]
A
Like, yes, that's exactly right.
- [52:26]
B
That's exactly right. Yeah. So I did want to quickly shout out too, by the way, if you're interested in sort of like the old, like, like older videos. And as, as you had mentioned, like, you know, he's, he's a creator that, that you've been following for a while. If you ever want to go through the entire history of AI video, a guy named Rich Klein has. Oh yeah, great website called AI Film History. And it like literally begins at the very beginning. And it's a very pretty website. You can just kind of scroll down, you see all of these examples. I love it. Exactly how far we've come in the short amount of time. It's pretty, it's. It's a great website, great resource.
- [52:59]
A
Tim, thank you so much for joining us this week. This was awesome and I hope our viewers listen to it. Go. What's the name of your YouTube challenge is Shout it out here at the end so people can go find it.
- [53:06]
B
Theoretically media, which has way too many syllables for, for anybody to, to type easily into, into a web browser. So yeah, theoretically media or just hit the link down below, which I'm sure will be there.
- [53:18]
A
So that's awesome. Thanks, Tim. We'll see you all next week. Kevin will be back.
- [53:21]
B
Bye.
- [53:21]
A
Bye.
- [53:22]
B
Bye.