Real-Time AI Video (FREE/Cheap!) & Seedance 2.5 Footage

[00:00:00]Well, things are heating up in the creative AI space. We've got a lot to cover today, including a big update in real time AI video, some seed dance seed news, a quick check-in with our friends over at Martini, going over their new camera control feature, and a really interesting new workflow that I think you're going to want to check out.

[00:00:20]Kicking off, Dart recently dropped Lucy 2.5. This is the latest rev of their realtime video model. Uh, this one edits video live at 1080p 30 frames a second with near zero latency. Now, as a quick note, uh, this language always bothers me. Uh, obviously this isn't actually editing the video in the traditional like nonlinear editing sense, but rather it's doing the work of like VFX or compositing on top of the video. Not to say that any of this is any less impressive. I just I know that there are a few of us out there that always kind of bristle when we hear video editing in this context. But again, Lucy 2.5 is pretty impressive as you can like swap a character, restyle the scene, drop in VFX or remove objects all midstream driven by text prompts and reference images. The way this works is what Deart is calling self anchoring. So the model adopts to its own generated output as the new reference point, which is how we avoid everything like devolving into bobbyac people or like in insane psychedelic nightmares. Granted, are there sane psychedelic nightmares? I I mean, I've never had one. If you have, let me know in the comments. In the meantime, let's go head over to Deart where they've set up a playground for us to try out Lucy 2.5. And actually, good news outside of the playground as well when it comes to pricing. So, yeah, it's always weird being on this camera. But, as we can see, uh I am now an animated character. Um there's a couple of presets down here as well, like uh hair on fire. Let's try that. Yeah. Hey. Wow.

[00:01:50]Uh I did need a haircut anyways, but now like where's Flamethrower Girl? Um, you can reference um other thing. That's actually the thing that interests me the most. Let's toss in Renfield here and say uh change the man to the attached reference character. See how this ends up looking. Um, so did that fire? Yeah, there it is. So, yeah, there I am. Now I'm Renfield. Um, pretty good. Uh, there's a little bit of blurring down here. What is Oh. Oh, you know what? That might be that might be because of the microphone, I wonder. Um, yeah, it's always interesting seeing uh how these characters react. He's a little like I don't know the uh I got to work on how you act in AI uh restylizations. So, currently the easiest way to access Lucy 2.5 is via this playground. Uh although it is again available via API, I did not see an app version of this yet. Uh that was the thing that was kind of showcased in the demo with this guy walking around with his phone and you know editing things on the fly. But in good news, although you know the app version doesn't exist, it looks like you might actually be able to build one since the API costs for this at least at 720 uh appear to be 2 cents a second. So I mean that's pretty great considering that generation costs for everything have been skyrocketing to be honest. You know costs for everything have been skyrocketing. uh 2 cents a second. That is refreshing news. So, I'll definitely be keeping an eye on this one. I do think that one like potential use case for Lucy 2.5 is to treat it like live previs where you can, you know, shoot something on your phone and then, you know, take it through a videotovideo model, say like obviously Cance 2.0 or 2.5 for a final render pass. And again, considering that the API is available, I mean, this is something that, you know, potentially someone brave enough out there might be able to tackle on their own. I mean, I'd give it a shot, but I am super low on Fable 5 tokens right now, and I have not gotten access to Kimmy K3 yet. Moving on, we are all still on Seance 2.5 watch. No release date as of yet, but samples uh kind of continue to trickle out. Here is the latest.

[00:04:30]I mean, overall, I think this is Pretty cool output. It was posted up over on Billy Billy, so a big meihow to our viewers over there. Um, I mean, again, overall, I kind of really like this output. I dig that like late '7s, early 80s vibe that it has going on. Very Blues Brothers. The Crash looks really good. Yeah, there is some water that sprays up there seemingly out of nowhere, but yeah, it kind of works. Um, you know, overall, I think it looks good. There are there is of course stuff to nitpick. I don't think I'm being overly critical, but there is like a phantom car that appears over here.

[00:05:02]Another point is that as these two cars crash, I mean, in in terms of like real world physics, obviously that car that was behind them would definitely crash into them as well. So, there is that as well. Again, a lot of this might be splitting hairs, but also might be the reason that it is delayed a bit longer.

[00:05:19]Look, if you think that this output looks amazing, that's fantastic. if you think it needed more work. Well, again, I mean, I think that's what the delay is about. Obviously, my eyes are peeled and I will let you know as soon as I hear anything about a release date. Worth a quick sidebar while we're in the sea dance section. Uh, director Neil Blamp of District 9 Elisium and Chappie fame, I guess. And well, actually to me the uh the greatest unmade alien movie. Well, he has released a short film created with CDS 2.0. Uh, you can watch it now.

[00:05:52]The link is down below. I also did want to let you guys know I am starting to hear a lot more about a seedance level model that is waiting in the wings. I mean, yes, on the one hand, it is kind of like a a no-brainer, but I have spoken with a few people who tell me that they have seen it. They can't really say much more than yes, they have seen it. It's very good, and it should be here soon. So yes, while you can file this news under nothing burger, I did just want to kind of put it out there as there's definitely been a lot more concern lately uh about bike dance essentially dominating everything and probably could use some competition which you know I mean it always helps with pricing. Speaking of bike dance completely crushing it in just about every area. Um we recently covered seed 1.0 the audio generator. Uh yeah it is very good. Interestingly Bike just released an update. It's not really an update. Like, this isn't a 1.1. It is still 1.0, but improvements have been made, including timing generations and the ability to pin dialogue to exact times. Uh, you know, basically more control. Take a quick listen and actually a look as well.

[00:07:00]>> Wait, why do I have an Indian accent?

[00:07:03]>> Isn't that what you wrote in the script?

[00:07:05]And you're going to get hit in 3 2 1.

[00:07:10]Then Hawk shows up and says, >> "Watch out.

[00:07:12]>> Gate 7 is closing. Bring him through."

[00:07:15]>> Copy. Don't look around. Just keep walking.

[00:07:20]>> Easy, man. We ain't missing this flight.

[00:07:24]>> I have to say, this feels way too real.

[00:07:27]>> It's only the beginning.

[00:07:28]>> Remember kids, sound is half your picture. Moving on, we're going to take a look at an experimental video generation workflow that I think you're going to find pretty interesting. But first, we're going to head over to Martini to check out some of their latest features. That part of the video is sponsored, but as always, Martini is super chill, which is also the best way to drink one. So, no hard sell here.

[00:07:49]Moving over to a visit with our friends at Martini, who were kind enough to sponsor today's video and I maybe buy you a drink. I'll I'll tell we'll talk about that in a little bit. Now, if you've been following the channel for a while, you know that I've covered Martini a few times in the past. And well, they have been busy at work with some new features for us, but the big recent one is camera motion. And this one kind of builds off of the step into set feature. Uh camera motion allows you to well literally control your camera movements in an imagetovideo generation.

[00:08:19]So, the way this works is pretty neat.

[00:08:21]Uh we have this image of flamethrower girl here standing in front of the meta building. Um we used this in a thumbnail uh a few videos back. um come up to your three dots up here and then um we want to step into set and then camera motion.

[00:08:33]From there it kind of opens up into like a gossian splat sort of thing. Um you'll notice that you can change your uh focal lengths up here, your lens length rather. Um and then we have its timeline down here. So if we if we if we scrub down the timeline a little bit, you can add a key frame in. So um let's go ahead and add a key. Well, hang on. Let's see.

[00:08:52]Let's move her around or move the camera around. So WSD keys move the camera around. And then your mouse um you know just kind of places where you want the camera. Um from here we just add a key frame in. And now you'll see the the key frame has appeared. We can actually slide the key frame down if we want to.

[00:09:08]Um if you grab a hold of it. Oh, it's little arrows right there. Um and then we have a uh like motion intensity. Some options on that. I just generally leave it on this one. Um but if we scroll back to the beginning, you'll see we're at the end. Um if you scroll back to the beginning, you'll see now we have our movement applied. Uh, and you can add multiple key frames in, too. So, if you wanted to, um, say get really wacky with it, uh, you can. So, um, add another key frame in really short. This is going to be a bit of a mess, but, uh, we'll see.

[00:09:38]So, um, as you can see, and then it should snap over that way. So, yeah, it obviously, you know, you just control the camera. Uh when you're happy with what you got um you just hit save camera move and then that will repopulate over here in which we have our motion control or motion camera uh over here along with our start frame. Um and then ingredients essentially you know elements um uh audio as well. You can throw in like a seed 1.0 audio file if you want to. And then prompt box down here. Obviously this is generating in uh Cance 2.0. So this is just kind of a silly test that I ended up putting together with that uh with camera movement that makes no sense. and a prompt that just says, "She's laughing.

[00:10:18][laughter] >> We're moving."

[00:10:22]>> What I like about this new feature is there's just a lot of different ways to approach it. For example, taking our uh crew cabin quarters uh from a spaceship.

[00:10:30]Uh I think we've used this one before.

[00:10:32]And we can add in our crew member as an element uh and then just, you know, have her walk in the room and sit down in the bed. Now, is this the greatest of camera moves for this particular shot? Not really. But um it just again it just sort of opens up a door to creating a workflow in which you are actually plotting out your camera move before your actors even step on the set. Um your virtual actors and your virtual set. And what's cool is that Martini is not even done cooking on this one yet.

[00:11:00]There's an upcoming ability to be able to use your phone as a motion controller as well. So that's pretty awesome. Uh, the other, and I'll admit kind of like not necessarily mind-blowing game-changing feature recently introduced, but also super necessary and I think an extremely helpful quality of life feature. Um, and again, particularly since Martini is canvas- based, is that they've introduced the uh the idea of a library. And look, I know it probably does not seem like that big a deal and I'm actually just getting started with putting stuff together here, but um you know, obviously we can put together a library for flamethrower girl and that means that any canvas that I later visit or any other project that I'm in in Martini, I can have access to all of my flamethrower girl media. Um so, you know, and obviously you can build that out for characters or locations or what whatever else you need. Uh, the other kind of cool part, if you're into the whole MCP thing, uh, you can just pop into something like Claude and say something like, uh, you know, hop into my library and place the image of Flamethrower Girl with her flamethrower on the canvas of a new project. Oh, look, it's another Claude update. It was shocking. And now my little minion has done its job. Uh, you know, I know this seems pretty stupid right now, but I think that when you start getting like building up a pretty substantial library, this could come in very handy. And the martini team is not even done. they're still shaking or or stirring. Uh so if you want to check them out, they have been kind enough to offer you guys viewers of this channel 50% off your first month of subscription. So just hit the link below and enjoy your martini. Moving on, I ran across a pretty new interesting technique from Oak. Essentially the idea here is to utilize depth maps as your storyboard uh layouts uh instead of like traditional like you know story traditional uh generated uh storyboard frames. So the idea here is that again uh you know we have a nine panel grid here uh that takes the place of depth maps and that's referencing your characters and I guess just one image that kind of defines the look. I I think this works pretty well. Uh, it definitely retains that kind of like 80s, like late 80s like dark fantasy kind of vibe to it. Very like Hensones, dark crystally. Uh, but also by way of um like Drizzit, I want to say dark elf uh sneaky archer kind of thing. Yeah, this is very cool. Now, I think it is important to note that these are not actual depth maps. These are well GPT images idea of depth maps. So much like we've seen when uh we give like nano banana or uh image one like you know make me a uh you know blender UI it's kind of there but it's kind of not. So these are not really truly depth maps.

[00:13:42]Um so it is hard to say if how much effect this has. So I was curious to see how this would perform with an actual depth map. Um and then additionally I remembered oh yeah I built a tool for that. Yeah. A few videos back I made theoretically pose V4. um and release that. Link to that will be down below.

[00:14:00]Actually, as I was going through this one, I for V5, one of the things I need to do is just uh be able to put in a still image and get a depth map from that. So, um regardless, uh taking this and then utilizing it essentially with the same prompt and same references, uh flamethrower girl and the location itself. Um and then just saying to reference video one as essentially the uh the depth reference. We're gonna I mean essentially I don't know what I was thinking but essentially it was it kind of looks exactly the same. It looks exactly the same until you compare them back to back. I'll fully admit it is uh definitely very subtle but I mean it's there. The characters in the foreground look much more separated from the background and you know again the overall color palette has changed I think to its benefit. Now I did want to continue on and experiment around more with this but Cance has been acting a little funky lately. um mostly likely due to uh the imminent release or testing or GPU allocation to Cance 2.5.

[00:15:01]So um yeah, I'm I'm going to bench this experiment um for 2.5, but I would say for sure uh I mean give the whole depth map thing a shot if you want. Uh the link to the pose control thing is down below along with the whole video that explains it. Final admin notes. Yes, I am aware I do need a haircut. I think this is going to be a pretty busy week, so I do not think that I'm going to have time to go out and get one. Mostly what I'm trying to say is I will see you again very soon. As always, I thank you for watching. by Davis T.