I Thought Local AI Video Was Good

This was generated on my computer. This was generated on my computer. >> Can you tell which one of me is the real >> Can you tell which one of me is the real >> Can you tell which one of me is the real one? one? one? >> And so was this. >> And so was this. >> And so was this. >> Does this support the new Phantom 5090? >> Does this support the new Phantom 5090? >> Does this support the new Phantom 5090? >> Oh, yeah, 24 lanes. >> Oh, yeah, 24 lanes. >> Oh, yeah, 24 lanes. >> You might have caught that video where I >> You might have caught that video where I >> You might have caught that video where I compared a bunch of quantizations, and I compared a bunch of quantizations, and I compared a bunch of quantizations, and I generated local videos using 1 and LTX. generated local videos using 1 and LTX. generated local videos using 1 and LTX. Start to finish on my machine. Nothing Start to finish on my machine. Nothing Start to finish on my machine. Nothing to send to the cloud. A few years ago, to send to the cloud. A few years ago, to send to the cloud. A few years ago, that was just science fiction. Now, it that was just science fiction. Now, it that was just science fiction. Now, it just runs locally, privately, and free. just runs locally, privately, and free. just runs locally, privately, and free. And for about 30 seconds, And for about 30 seconds, And for about 30 seconds, well, in this case, at most 10 seconds. well, in this case, at most 10 seconds. well, in this case, at most 10 seconds. That's the whole story. It's amazing That's the whole story. It's amazing That's the whole story. It's amazing that it works at all, which is exactly that it works at all, which is exactly that it works at all, which is exactly what we thought of LLMs a few years ago. what we thought of LLMs a few years ago. what we thought of LLMs a few years ago. Would I actually use this? Would I actually use this? Would I actually use this? >> Did you try turning it off and on again? >> Did you try turning it off and on again? >> Did you try turning it off and on again? >> I >> I >> I >> Because there are some issues. So, I >> Because there are some issues. So, I >> Because there are some issues. So, I took the exact same prompts and ran them took the exact same prompts and ran them took the exact same prompts and ran them through one of the best frontier video through one of the best frontier video through one of the best frontier video models out there right now. Same ideas, models out there right now. Same ideas, models out there right now. Same ideas, but not the same result. Ho ho, not even but not the same result. Ho ho, not even but not the same result. Ho ho, not even close. And then, uh the yellow question close. And then, uh the yellow question close. And then, uh the yellow question comes up, because this kind of feels comes up, because this kind of feels comes up, because this kind of feels like where code was a couple years ago. like where code was a couple years ago. like where code was a couple years ago. Do you remember? First, AI could Do you remember? First, AI could Do you remember? First, AI could generate code, and we were so happy that generate code, and we were so happy that generate code, and we were so happy that we can chat with a model like ChatGPT, we can chat with a model like ChatGPT, we can chat with a model like ChatGPT, and it'll spit out some code for us. And and it'll spit out some code for us. And and it'll spit out some code for us. And now, it became agents that could now, it became agents that could now, it became agents that could understand the project, make changes, understand the project, make changes, understand the project, make changes, fix mistakes, and just keep going. So, fix mistakes, and just keep going. So, fix mistakes, and just keep going. So, is video about to go through the same is video about to go through the same is video about to go through the same shift? Not just prompt in, clip out, but shift? Not just prompt in, clip out, but shift? Not just prompt in, clip out, but something that understands what you're something that understands what you're something that understands what you're trying to make, figures out the pieces, trying to make, figures out the pieces, trying to make, figures out the pieces, and builds the next version. And that is and builds the next version. And that is and builds the next version. And that is where this got way more interesting.

where this got way more interesting. where this got way more interesting. >> Did you try turning it off and on again? >> Did you try turning it off and on again? >> Did you try turning it off and on again? >> So, first of all, this is 1 2.2, and >> So, first of all, this is 1 2.2, and >> So, first of all, this is 1 2.2, and it's doing a pretty decent job. This is it's doing a pretty decent job. This is it's doing a pretty decent job. This is the BF16 quality, so the highest. But, the BF16 quality, so the highest. But, the BF16 quality, so the highest. But, this is a very simple shot. Now, when this is a very simple shot. Now, when this is a very simple shot. Now, when you use a different model, like I'm you use a different model, like I'm you use a different model, like I'm using here, this is uh Seed Dance 2.0, using here, this is uh Seed Dance 2.0, using here, this is uh Seed Dance 2.0, and I'm accessing it through Higgs and I'm accessing it through Higgs and I'm accessing it through Higgs Field. Even if you give the same seed to Field. Even if you give the same seed to Field. Even if you give the same seed to a different model, it's going to look a different model, it's going to look a different model, it's going to look very different. And this looks really very different. And this looks really very different. And this looks really good. Seed Dance is actually one of the good. Seed Dance is actually one of the good. Seed Dance is actually one of the best, if not the best model right now. best, if not the best model right now. best, if not the best model right now. So, I decided to take a screenshot from So, I decided to take a screenshot from So, I decided to take a screenshot from the original video that I generated with the original video that I generated with the original video that I generated with 1 and LTX, and run it through Seed Dance 1 and LTX, and run it through Seed Dance 1 and LTX, and run it through Seed Dance to see what it could do. Wow. Now, that is impressive. It's got Wow. Now, that is impressive. It's got camera motion. camera motion. camera motion. It's got a slight zoom and the water It's got a slight zoom and the water It's got a slight zoom and the water surface is just jiggling a little bit. surface is just jiggling a little bit. surface is just jiggling a little bit. Very nice. Let's do something more Very nice. Let's do something more Very nice. Let's do something more complicated. Here's a lady walking complicated. Here's a lady walking complicated. Here's a lady walking through the forest. Now this is the through the forest. Now this is the through the forest. Now this is the highest quality I could generate with highest quality I could generate with highest quality I could generate with WEN and it's pretty good if you just WEN and it's pretty good if you just WEN and it's pretty good if you just watching it, but there are a couple of watching it, but there are a couple of watching it, but there are a couple of issues with consistency if you go issues with consistency if you go issues with consistency if you go frame-by-frame. Generally though, this frame-by-frame. Generally though, this frame-by-frame. Generally though, this is acceptable for most applications. is acceptable for most applications. is acceptable for most applications. However, look at this version that Seed However, look at this version that Seed However, look at this version that Seed Dance 2.0 gave me. That is impressive.

Dance 2.0 gave me. That is impressive. Dance 2.0 gave me. That is impressive. The hair bounces there. Her face is so The hair bounces there. Her face is so The hair bounces there. Her face is so much sharper. She's got a bounce to her much sharper. She's got a bounce to her much sharper. She's got a bounce to her step. It just looks realistic. There's step. It just looks realistic. There's step. It just looks realistic. There's even some camera motion going on. Now even some camera motion going on. Now even some camera motion going on. Now even the best model can't handle certain even the best model can't handle certain even the best model can't handle certain things like physics. There's some things things like physics. There's some things things like physics. There's some things with physics that is an issue like this with physics that is an issue like this with physics that is an issue like this marble rolling down a block of wood. marble rolling down a block of wood. marble rolling down a block of wood. This is WEN. You can see it's totally This is WEN. You can see it's totally This is WEN. You can see it's totally unrealistic. The two balls merge and unrealistic. The two balls merge and unrealistic. The two balls merge and Seed Dance even though the marble is Seed Dance even though the marble is Seed Dance even though the marble is really nice looking, really nice looking, really nice looking, the physics is just the physics is just the physics is just not working out here. It did add sound not working out here. It did add sound not working out here. It did add sound though. Speaking of sound, LTX is what I ran Speaking of sound, LTX is what I ran locally to get audio with the video locally to get audio with the video locally to get audio with the video synced up. Here is that tech support synced up. Here is that tech support synced up. Here is that tech support scene. scene. scene. >> Did you try turning it off and on again? >> I AM THE IT DEPARTMENT. >> I AM THE IT DEPARTMENT. >> IT'S PRETTY FUNNY, BUT there's >> IT'S PRETTY FUNNY, BUT there's >> IT'S PRETTY FUNNY, BUT there's definitely some issues with consistency. definitely some issues with consistency. definitely some issues with consistency. There's weird facial glitches going on There's weird facial glitches going on There's weird facial glitches going on and it's got this kind of rubbery feel and it's got this kind of rubbery feel and it's got this kind of rubbery feel to it. Now for Seed Dance, I uploaded to it. Now for Seed Dance, I uploaded to it. Now for Seed Dance, I uploaded just one frame. I believe it was one of just one frame. I believe it was one of just one frame. I believe it was one of these last frames and asked it to do the these last frames and asked it to do the these last frames and asked it to do the whole video and check this out.

whole video and check this out. whole video and check this out. >> Did you try turning it off and on again? >> I AM THE IT DEPARTMENT. >> I AM THE IT DEPARTMENT. >> That is crazy good. First of all, their >> That is crazy good. First of all, their >> That is crazy good. First of all, their voices are just really realistic. He's voices are just really realistic. He's voices are just really realistic. He's got a bit of a sweaty forehead from got a bit of a sweaty forehead from got a bit of a sweaty forehead from working on this IT problem that's working on this IT problem that's working on this IT problem that's bothering him. We all know what that's bothering him. We all know what that's bothering him. We all know what that's like and look at his reaction as soon as like and look at his reaction as soon as like and look at his reaction as soon as she says that to him. she says that to him. she says that to him. Just uh, you know, to push his buttons a Just uh, you know, to push his buttons a Just uh, you know, to push his buttons a bit and it worked obviously. But every bit and it worked obviously. But every bit and it worked obviously. But every single frame in this video is single frame in this video is single frame in this video is consistent. There's no weird glitches consistent. There's no weird glitches consistent. There's no weird glitches with their faces. They got that nailed. with their faces. They got that nailed. with their faces. They got that nailed. Here's one that started from an actual Here's one that started from an actual Here's one that started from an actual image of one of my videos with Dan at image of one of my videos with Dan at image of one of my videos with Dan at Micro Center. This is LTX and this is Micro Center. This is LTX and this is Micro Center. This is LTX and this is the first frame. It looks fine. the first frame. It looks fine. the first frame. It looks fine. >> Does this support the new Phantom 5090? >> Does this support the new Phantom 5090? >> Does this support the new Phantom 5090? Oh, yeah. 24 lanes. >> But then it kind of drifts off a little >> But then it kind of drifts off a little bit. bit. bit. Wait. THEY LIED TO US AGAIN. THEY LIED TO US AGAIN. >> I LOVE THAT. BUT if you take a closer >> I LOVE THAT. BUT if you take a closer >> I LOVE THAT. BUT if you take a closer look, like what's going on with the face look, like what's going on with the face look, like what's going on with the face over there? What's going on with Dan's over there? What's going on with Dan's over there? What's going on with Dan's face? Not realistic. So I had Seed Dance face? Not realistic. So I had Seed Dance face? Not realistic. So I had Seed Dance 2.0 do it. The original is 10 seconds 2.0 do it. The original is 10 seconds 2.0 do it. The original is 10 seconds and if I gave it that whole script to do and if I gave it that whole script to do and if I gave it that whole script to do in a shorter amount of time, it had to in a shorter amount of time, it had to in a shorter amount of time, it had to kind of compress it. Here's what I got.

kind of compress it. Here's what I got. kind of compress it. Here's what I got. >> Does this support the new Phantom 5090? >> Does this support the new Phantom 5090? >> Does this support the new Phantom 5090? Oh, yeah. 24 lanes. WAIT. Oh, yeah. 24 lanes. WAIT. Oh, yeah. 24 lanes. WAIT. THEY LIED TO US AGAIN. THEY LIED TO US AGAIN. THEY LIED TO US AGAIN. >> [laughter] >> [laughter] >> [laughter] >> IT DID A GOOD JOB. I DON'T KNOW. It it >> IT DID A GOOD JOB. I DON'T KNOW. It it >> IT DID A GOOD JOB. I DON'T KNOW. It it literally made Dan red with rage. That's literally made Dan red with rage. That's literally made Dan red with rage. That's not supposed to be a literal thing. So not supposed to be a literal thing. So not supposed to be a literal thing. So there's some downsides here, but the there's some downsides here, but the there's some downsides here, but the consistency is incredible with both of consistency is incredible with both of consistency is incredible with both of our faces, even when he's really mad. So our faces, even when he's really mad. So our faces, even when he's really mad. So local AI is pretty impressive already, local AI is pretty impressive already, local AI is pretty impressive already, but as soon as you put it next to the but as soon as you put it next to the but as soon as you put it next to the frontier model here, Seed Dance 2.0. Not frontier model here, Seed Dance 2.0. Not frontier model here, Seed Dance 2.0. Not all of them are good by the way, but all of them are good by the way, but all of them are good by the way, but that happens to be a really good one. that happens to be a really good one. that happens to be a really good one. The realism gap just shows up. So the The realism gap just shows up. So the The realism gap just shows up. So the question is would I use clips of the question is would I use clips of the question is would I use clips of the Seed Dance generated one? I would say Seed Dance generated one? I would say Seed Dance generated one? I would say yeah. And actually, I have used it in a yeah. And actually, I have used it in a yeah. And actually, I have used it in a couple of videos recently. Some of you couple of videos recently. Some of you couple of videos recently. Some of you even asked me if the tomato scene was even asked me if the tomato scene was even asked me if the tomato scene was real or not. Not that I would ever real or not. Not that I would ever real or not. Not that I would ever actually do that on my desk in front of actually do that on my desk in front of actually do that on my desk in front of my computers. my computers. my computers. >> Woah. >> Woah. >> Woah. >> I still have faith though in local >> I still have faith though in local >> I still have faith though in local models. They are improving as we go and models. They are improving as we go and models. They are improving as we go and they're a little bit behind, but they'll they're a little bit behind, but they'll they're a little bit behind, but they'll get there. It's also cheap to run local get there. It's also cheap to run local get there. It's also cheap to run local models. First of all, the models weights models. First of all, the models weights models. First of all, the models weights are free and second, it costs you are free and second, it costs you are free and second, it costs you electricity and you can run it over and electricity and you can run it over and electricity and you can run it over and over and over again until you get the over and over again until you get the over and over again until you get the best result. Whereas something like best result. Whereas something like best result. Whereas something like Higgs Field, you have to pay And you Higgs Field, you have to pay And you Higgs Field, you have to pay And you have to watch your credits. I've been have to watch your credits. I've been have to watch your credits. I've been subscribed to the Ultra plan for a subscribed to the Ultra plan for a subscribed to the Ultra plan for a couple months now, and it's halfway couple months now, and it's halfway couple months now, and it's halfway through the month, and I'm halfway through the month, and I'm halfway through the month, and I'm halfway through my credits. So, I'm doing okay.

through my credits. So, I'm doing okay. through my credits. So, I'm doing okay. But, in the back of my mind, I'm always But, in the back of my mind, I'm always But, in the back of my mind, I'm always thinking about it. So, what I would do thinking about it. So, what I would do thinking about it. So, what I would do is I would experiment locally as much as is I would experiment locally as much as is I would experiment locally as much as you want, and then when you have you want, and then when you have you want, and then when you have something kind of flushed out, when you something kind of flushed out, when you something kind of flushed out, when you have a good prompt, then you hit up have a good prompt, then you hit up have a good prompt, then you hit up Higgsfield and one of the models they Higgsfield and one of the models they Higgsfield and one of the models they have available there. By the way, if you have available there. By the way, if you have available there. By the way, if you go to video, you have not only Seed go to video, you have not only Seed go to video, you have not only Seed Dance 2, but you have Kling 3, Motion Dance 2, but you have Kling 3, Motion Dance 2, but you have Kling 3, Motion Control, Sora, Vio, all those available Control, Sora, Vio, all those available Control, Sora, Vio, all those available to you, even 1 2.7. But, there's also to you, even 1 2.7. But, there's also to you, even 1 2.7. But, there's also something new. And this is something I something new. And this is something I something new. And this is something I haven't been able to reproduce yet haven't been able to reproduce yet haven't been able to reproduce yet locally is the agentic workflows with locally is the agentic workflows with locally is the agentic workflows with supercomputer that they have here. This supercomputer that they have here. This supercomputer that they have here. This is brand new. So, I tried to hook up is brand new. So, I tried to hook up is brand new. So, I tried to hook up Higgsfield CLI. They have both now, MCP Higgsfield CLI. They have both now, MCP Higgsfield CLI. They have both now, MCP and CLI. I showed off the MCP in my and CLI. I showed off the MCP in my and CLI. I showed off the MCP in my previous workflow. But, I tried the new previous workflow. But, I tried the new previous workflow. But, I tried the new CLI, and it works for generating clips, CLI, and it works for generating clips, CLI, and it works for generating clips, for generating images. It's actually for generating images. It's actually for generating images. It's actually really good. But, when it comes to really good. But, when it comes to really good. But, when it comes to stitching things together in agentic stitching things together in agentic stitching things together in agentic workflows, that's where supercomputer workflows, that's where supercomputer workflows, that's where supercomputer comes in. And right now, it's still in comes in. And right now, it's still in comes in. And right now, it's still in the UI. At first, you go in here, and the UI. At first, you go in here, and the UI. At first, you go in here, and it's like, "Oh, this is just another it's like, "Oh, this is just another it's like, "Oh, this is just another chat box." But, not really. If you take chat box." But, not really. If you take chat box." But, not really. If you take a closer look, you have access not only a closer look, you have access not only a closer look, you have access not only to Opus and Opus 4.8, 4.6, Sonnet. You to Opus and Opus 4.8, 4.6, Sonnet. You to Opus and Opus 4.8, 4.6, Sonnet. You have Gemini in here, Gemini Pro, Gemini have Gemini in here, Gemini Pro, Gemini have Gemini in here, Gemini Pro, Gemini Flash, OpenAI's GPT 5.5, Grok is in Flash, OpenAI's GPT 5.5, Grok is in Flash, OpenAI's GPT 5.5, Grok is in here. And you have some open weight here. And you have some open weight here. And you have some open weight models like Kimmy K 2.6 and Deep Seek.

models like Kimmy K 2.6 and Deep Seek. models like Kimmy K 2.6 and Deep Seek. Everything is accessible through here. Everything is accessible through here. Everything is accessible through here. But, why would you need that if you But, why would you need that if you But, why would you need that if you already have like a Claude subscription? already have like a Claude subscription? already have like a Claude subscription? Well, first of all, you have access to Well, first of all, you have access to Well, first of all, you have access to all these, but this understands your all these, but this understands your all these, but this understands your videos, and it understands how to work videos, and it understands how to work videos, and it understands how to work with videos and build things into with videos and build things into with videos and build things into products. So, I'm going to take it for a products. So, I'm going to take it for a products. So, I'm going to take it for a test drive doing something that's maybe test drive doing something that's maybe test drive doing something that's maybe useful to me, but basically, the sky's useful to me, but basically, the sky's useful to me, but basically, the sky's the limit here. Now, sometimes people the limit here. Now, sometimes people the limit here. Now, sometimes people complain that I have a wrinkly shirt in complain that I have a wrinkly shirt in complain that I have a wrinkly shirt in my videos. Okay, I don't take offense. my videos. Okay, I don't take offense. my videos. Okay, I don't take offense. It's okay. You can make fun of me. It's okay. You can make fun of me. It's okay. You can make fun of me. >> [laughter] >> I like to laugh at myself. It's part of >> I like to laugh at myself. It's part of life. Don't take yourself too seriously. life. Don't take yourself too seriously. life. Don't take yourself too seriously. However, this gave me an idea. What if I However, this gave me an idea. What if I However, this gave me an idea. What if I took a clip of one of my videos and took a clip of one of my videos and took a clip of one of my videos and asked Hicksfield supercomputer to asked Hicksfield supercomputer to asked Hicksfield supercomputer to unwrinkle my shirt. Found this clip. unwrinkle my shirt. Found this clip. unwrinkle my shirt. Found this clip. Nobody actually tells you what you gave Nobody actually tells you what you gave Nobody actually tells you what you gave up to get there. And I thought, okay, up to get there. And I thought, okay, up to get there. And I thought, okay, you know what? The shirt is not that you know what? The shirt is not that you know what? The shirt is not that wrinkly. Come on. And maybe I want to do wrinkly. Come on. And maybe I want to do wrinkly. Come on. And maybe I want to do something a little bit more drastic. something a little bit more drastic. something a little bit more drastic. Like, I don't know, wear a tuxedo. I Like, I don't know, wear a tuxedo. I Like, I don't know, wear a tuxedo. I just wanted to see what it can do. So, I just wanted to see what it can do. So, I just wanted to see what it can do. So, I asked it to clean up and get rid of some asked it to clean up and get rid of some asked it to clean up and get rid of some of the wrinkles, but maintain my exact of the wrinkles, but maintain my exact of the wrinkles, but maintain my exact performance, my face, everything about performance, my face, everything about performance, my face, everything about the image. And it thought about it for a the image. And it thought about it for a the image. And it thought about it for a while, but it ended up recommending that while, but it ended up recommending that while, but it ended up recommending that I don't use a model, which is actually I don't use a model, which is actually I don't use a model, which is actually pretty cool. It found that it's not pretty cool. It found that it's not pretty cool. It found that it's not capable of doing what I asked it to do, capable of doing what I asked it to do, capable of doing what I asked it to do, and it suggested some other workflows.

and it suggested some other workflows. and it suggested some other workflows. Like, you should use DaVinci Resolve Like, you should use DaVinci Resolve Like, you should use DaVinci Resolve workflow to do it yourself less than 5 workflow to do it yourself less than 5 workflow to do it yourself less than 5 minutes. Or you can use After Effects to minutes. Or you can use After Effects to minutes. Or you can use After Effects to draw a soft mask around the shirt torso. draw a soft mask around the shirt torso. draw a soft mask around the shirt torso. That's smart, though. I'm I don't have That's smart, though. I'm I don't have That's smart, though. I'm I don't have those kinds of skills. I'm not very good those kinds of skills. I'm not very good those kinds of skills. I'm not very good at After Effects, so I'm not going to do at After Effects, so I'm not going to do at After Effects, so I'm not going to do that. I want the easy way out, all that. I want the easy way out, all that. I want the easy way out, all right? So, instead of that, I asked it right? So, instead of that, I asked it right? So, instead of that, I asked it to do a costume change for me. I gave it to do a costume change for me. I gave it to do a costume change for me. I gave it my video clip, asked it to pick a frame, my video clip, asked it to pick a frame, my video clip, asked it to pick a frame, the best frame, so it needed to analyze the best frame, so it needed to analyze the best frame, so it needed to analyze my video, which it can do, and instead my video, which it can do, and instead my video, which it can do, and instead of the blue shirt, have me wear a of the blue shirt, have me wear a of the blue shirt, have me wear a tuxedo. It ran through the analysis, and tuxedo. It ran through the analysis, and tuxedo. It ran through the analysis, and this is something that its native agent this is something that its native agent this is something that its native agent can do. All that stuff is built in to can do. All that stuff is built in to can do. All that stuff is built in to the knowledge of Hicksfield the knowledge of Hicksfield the knowledge of Hicksfield supercomputer. So, you don't need to supercomputer. So, you don't need to supercomputer. So, you don't need to worry about training it, downloading the worry about training it, downloading the worry about training it, downloading the right skills, any of that. It ended up right skills, any of that. It ended up right skills, any of that. It ended up finding the best frame. It even showed finding the best frame. It even showed finding the best frame. It even showed me the options that it considered, and me the options that it considered, and me the options that it considered, and then it generated this image out of the then it generated this image out of the then it generated this image out of the best frame, which is me in a tuxedo in best frame, which is me in a tuxedo in best frame, which is me in a tuxedo in my own office. Now, since this has my own office. Now, since this has my own office. Now, since this has access to all the best models out there, access to all the best models out there, access to all the best models out there, it decides what model it wants to use. it decides what model it wants to use. it decides what model it wants to use. In this case, it used GPT 2.0 image, In this case, it used GPT 2.0 image, In this case, it used GPT 2.0 image, which in my opinion is the best model to which in my opinion is the best model to which in my opinion is the best model to edit images. Keeps everything consistent edit images. Keeps everything consistent edit images. Keeps everything consistent except what you asked it to edit. So, except what you asked it to edit. So, except what you asked it to edit. So, then I said, "Use the edited image as then I said, "Use the edited image as then I said, "Use the edited image as the start frame and generate a short 3-5 the start frame and generate a short 3-5 the start frame and generate a short 3-5 second realistic talking head video."

second realistic talking head video." second realistic talking head video." Now, it did that. However, it didn't end Now, it did that. However, it didn't end Now, it did that. However, it didn't end up being what I wanted. Welcome back to up being what I wanted. Welcome back to up being what I wanted. Welcome back to another video. Today, we're going to another video. Today, we're going to another video. Today, we're going to take a closer look at Although, I must take a closer look at Although, I must take a closer look at Although, I must say, it's very James Bondy, so it's very say, it's very James Bondy, so it's very say, it's very James Bondy, so it's very appropriate that it gave me a British appropriate that it gave me a British appropriate that it gave me a British accent in this case. But, that's not accent in this case. But, that's not accent in this case. But, that's not what I wanted. I wanted it to use my what I wanted. I wanted it to use my what I wanted. I wanted it to use my voice, my clip, just replace my clothes. voice, my clip, just replace my clothes. voice, my clip, just replace my clothes. Now, HeyGen Field has a few different Now, HeyGen Field has a few different Now, HeyGen Field has a few different workflows to be able to handle this. And workflows to be able to handle this. And workflows to be able to handle this. And I can go and do this manually using all I can go and do this manually using all I can go and do this manually using all their tools like lip sync. They give you their tools like lip sync. They give you their tools like lip sync. They give you access to Kling, of course, so you can access to Kling, of course, so you can access to Kling, of course, so you can do that using Kling edit. But, I wanted do that using Kling edit. But, I wanted do that using Kling edit. But, I wanted to see if I can just prompt the agent to to see if I can just prompt the agent to to see if I can just prompt the agent to do it because that's what it's for. So, do it because that's what it's for. So, do it because that's what it's for. So, in the same context, I gave it my video in the same context, I gave it my video in the same context, I gave it my video again and the new screenshot of me in again and the new screenshot of me in again and the new screenshot of me in the tux. And I said, "Use the audio from the tux. And I said, "Use the audio from the tux. And I said, "Use the audio from this video and generate a new video with this video and generate a new video with this video and generate a new video with the guy in the tuxedo." So, it has to the guy in the tuxedo." So, it has to the guy in the tuxedo." So, it has to figure out a few things now. It has to figure out a few things now. It has to figure out a few things now. It has to first figure out how to get the audio first figure out how to get the audio first figure out how to get the audio from the video, then use that audio to from the video, then use that audio to from the video, then use that audio to drive the lip synced tuxedo guy and drive the lip synced tuxedo guy and drive the lip synced tuxedo guy and submit all that to the correct model. submit all that to the correct model. submit all that to the correct model. And the one correct model that can do And the one correct model that can do And the one correct model that can do that right now is C Dance 2.0. There it that right now is C Dance 2.0. There it that right now is C Dance 2.0. There it goes.

goes. goes. It It figured out the length of the It It figured out the length of the It It figured out the length of the video, the length of the audio, cut the video, the length of the audio, cut the video, the length of the audio, cut the audio out using FFmpeg that it itself is audio out using FFmpeg that it itself is audio out using FFmpeg that it itself is running up in the cloud, and then it running up in the cloud, and then it running up in the cloud, and then it figured out that it needs to use C figured out that it needs to use C figured out that it needs to use C Dance. So, it submits the image and it Dance. So, it submits the image and it Dance. So, it submits the image and it submits the audio. And there it is. It submits the audio. And there it is. It submits the audio. And there it is. It generated the video. Nobody actually generated the video. Nobody actually generated the video. Nobody actually tells you what you gave up to get there. tells you what you gave up to get there. tells you what you gave up to get there. >> [laughter] >> [laughter] >> [laughter] >> That is incredible. So, this whole >> That is incredible. So, this whole >> That is incredible. So, this whole agentic workflow feels like cursor, but agentic workflow feels like cursor, but agentic workflow feels like cursor, but for video and media. Not everything is for video and media. Not everything is for video and media. Not everything is perfect, though. I did actually run into perfect, though. I did actually run into perfect, though. I did actually run into a couple of issues where a couple of issues where a couple of issues where I was trying to render that woman in the I was trying to render that woman in the I was trying to render that woman in the forest, but it it gave me a not suitable forest, but it it gave me a not suitable forest, but it it gave me a not suitable for work and it wouldn't do it. But, for work and it wouldn't do it. But, for work and it wouldn't do it. But, after I retried it, it did it. So, yeah, after I retried it, it did it. So, yeah, after I retried it, it did it. So, yeah, that's another difference between doing that's another difference between doing that's another difference between doing it locally where it always works and it locally where it always works and it locally where it always works and doing it on the server where there are doing it on the server where there are doing it on the server where there are some checks and it's going to limit what some checks and it's going to limit what some checks and it's going to limit what you can generate. It's probably not you can generate. It's probably not you can generate. It's probably not going to do logos or famous people just going to do logos or famous people just going to do logos or famous people just for safety and legal reasons. So, if you for safety and legal reasons. So, if you for safety and legal reasons. So, if you care about control, local AI video is care about control, local AI video is care about control, local AI video is still pretty exciting and it's getting still pretty exciting and it's getting still pretty exciting and it's getting better all the time. If you care about better all the time. If you care about better all the time. If you care about realism and usable output today, the realism and usable output today, the realism and usable output today, the frontier models are clearly ahead. And frontier models are clearly ahead. And frontier models are clearly ahead. And if agents like Higgs Field supercomputer if agents like Higgs Field supercomputer if agents like Higgs Field supercomputer keep improving, the next competition keep improving, the next competition keep improving, the next competition isn't local versus cloud. It's going to isn't local versus cloud. It's going to isn't local versus cloud. It's going to be model versus workflow. Now, if you be model versus workflow. Now, if you be model versus workflow. Now, if you want to see how to get Claude code and want to see how to get Claude code and want to see how to get Claude code and Claude generating images and videos for Claude generating images and videos for Claude generating images and videos for you, you, you, >> [music] >> [music] >> [music] >> watch this video right here. Thanks for >> watch this video right here. Thanks for >> watch this video right here. Thanks for watching and I'll see you next time.