Full transcript
0:00Okay, so today I'm going to show you
0:00guys how to get the open-source
0:02Higgsfield. There are several ways you
0:03can do it, but this is kind of how it
0:05looks when you get it going. This is the
0:07whole system right here. It's pretty
0:10cool. What you can do Okay, so this is
0:12where it all comes from. It's comes from
0:13a Neil Matcha. The jet It's called The
0:15project is called open- generative-ai,
0:18and you have all of the files right
0:20here. So, if you want, you can download
0:22it yourself, and you can edit it however
0:24you want. There are some parts that
0:26don't work perfectly, and that's because
0:28it's an open-source project. That's kind
0:29of how this works. It's like generated
0:30by some person, and they made it for
0:32free, and they're giving this out for
0:33you to use and try. Okay, so there's two
0:36ways that you can use this open-source
0:37Higgsfield. So, we have the pay-per-use
0:40version and the free version. I'll walk
0:43through both of them, and I'll give you
0:44the benefits and the drawbacks. Let's do
0:46this. Okay, so first I'm going to show
0:48you the paid version. So, basically the
0:49paid version is super easy to get. All
0:51you have to do is you go to the hosted
0:53version, which is this website right
0:55here on my screen. You click on that,
0:58and it brings you straight to this thing
1:01that I was showing you earlier. Then
1:02from there, this allows you to use every
1:04single model in existence I can think
1:07of. I I'm sure there's more models out
1:09there, but there's just a ton on here,
1:11and there's lots of options, and it runs
1:13really well because it's running on
1:14their server. But, basically what you do
1:16for this one is it is the paid version,
1:18but it is a lot cheaper than using
1:20something like Higgsfield. Like this is
1:21the open-source version, so this is kind
1:23of like you have like the tiers of
1:25Higgsfield AI is very expensive, then
1:27you have middle, which is kind of middle
1:29of the road, which is this one, then you
1:31have the fully open-source where you
1:33download on your own computer, and that
1:34is free, basically. So, I'll show you
1:37how the middle tier works, and then
1:38we'll get into the free one. Okay,
1:39middle tier, what you're going to want
1:40to do is you will have to generate an
1:42API key. This is a Mu API. I don't know
1:44how to pronounce this properly.
1:45Basically, what you're going to do is
1:46generate an API key in here, and it's
1:48like a mega API key. And so, it covers
1:51every single model you can think in
1:53existence. A lot of them are cheaper
1:55than doing it through the actual
1:57company. So, that's nice. But, you have
1:59access to tons and tons of models. So,
2:01if you get really good with a certain
2:02model that's super cheap, you can
2:04generate a ton of stuff for a lot
2:06cheaper than it would be if you were to
2:07generate it like with Midjourney or or
2:10other companies that host it on their
2:12servers. So, it's a lot cheaper in that
2:14way. So, it's not completely free, but
2:16it is a lot cheaper, which is pretty
2:17nice. But, basically you have all these
2:19options and how it works is it connects
2:21with this API key. You just generate the
2:23API key. I I loaded $10 on there and
2:26I've been generating a ton of stuff. And
2:28that's when you have this section right
2:29here. So, you're able to basically
2:31generate anything through Nana Banana,
2:33Flex Dev, and there's tons of options,
2:35too. You can generate four at a time.
2:37You can change the aspect ratio. You can
2:38add reference images. There's just a ton
2:40of things that you can do. That's for
2:42the image stuff. Same thing with the
2:44video. You have all the options. You
2:45have like a bajillion options. And you
2:48can even download do the open-source
2:49models. They do host them on their own
2:51server, but the thing I'll show you
2:53later is you can host these models also
2:54on your own computer and these videos
2:57for free. I'll get to that. I'll get to
2:58that. Um there also are several models
3:00that you cannot host on your own
3:02computer, and that's like Seedance and
3:04Kling and VEO. These are all hosted on
3:07big fat servers cuz they take a lot to
3:09generate video. But, you're able to do
3:11all those at a more affordable price
3:14here in the video section. So, you can
3:16change the aspect ratio, change the
3:17time. You can change the quality, add
3:19other images. Like, there's a lot of
3:21stuff you can do. You also have this lip
3:23sync section where you can make it so it
3:24matches up with the voice of a face and
3:27you can throw an audio. Pretty simple.
3:29Then, you have Cinema Studio. This is
3:30where you can generate frame-by-frame
3:33and kind of like build the stuff
3:34together. And then, you have Marketing
3:36Studio, which is where you can build out
3:37ads for it. This is also much more
3:39affordable cuz you can choose the model
3:41and you can choose the variant how you
3:42want it to look, the avatar. Like,
3:44there's a ton of cool options in here.
3:46Then, also in here you do have workflows
3:48where you can do certain things which
3:49are like more specialized. So, you do
3:51social media, e-commerce, fashion,
3:52featured, featured social media, like
3:55all these things are are things that can
3:57help you like build better things and
3:59you can build your own workflows as well
4:00and community stuff which is generated
4:03from the community, which is cool. I
4:04like that it's all kind of centralized
4:05into one system. It's like basically the
4:07community building their own version of
4:09Higgsfield that's kind of all combined
4:10here. So, you don't have to pay for
4:12Higgsfield prices. So, you just pay like
4:14the base price and then everybody
4:16contributes to everybody. It's It's a
4:17pretty cool little system. It's not
4:19perfect though. It's not perfect. I'm
4:20not going to act like this is completely
4:21free cuz it's not. There is a free way
4:23to do it. It's not this one exactly.
4:25But, this is the open-source version.
4:26So, like I guess in the sense of
4:27open-source, like anybody can contribute
4:29to this platform right here and help
4:31build it to make it better, which is
4:33pretty cool. You also do have agents
4:35right here where you can literally just
4:37like like look at this. This is crazy.
4:39Like a veggie talk creator. Like there's
4:41a ton of different agents they can use
4:42to make content, which is pretty cool.
4:44All of these are you will have to pay
4:45for the the tokens, but then you also
4:48have apps where you can literally just
4:49build out your own template and create
4:51your own, I guess, wrapper in the sense
4:54of like you can create your own AI video
4:57studio and sell that to other people,
4:58which is pretty cool. And this one you
5:00can just like straight-up go to the
5:01GitHub and do it from there and build it
5:03out. And this is completely free. You
5:05would have to pay for the tokens, but if
5:06you're charging somebody like twice as
5:07much for the tokens, then it can make
5:09sense and it can be an actual business,
5:10which is pretty cool. All these GitHubs
5:12are right there and you can grab the
5:13templates and just download them and
5:15create your own AI business just like
5:18that. It's pretty gnarly how quick and
5:19easy it is. That is the main stuff on
5:22the hosted version, all right? Now, I
5:24know you guys didn't come here for the
5:25hosted version. I just want to show that
5:27this is kind of what the creator is
5:28making. It's like a cheaper and
5:30open-source alternative to doing
5:32Higgsfield, but there is like the other
5:34part where you can just host it on your
5:35own computer. So, let me show you kind
5:37of what that looks like. So, you're
5:38going to go into here, right? Go back to
5:40the main GitHub where I was. And this is
5:43where you can get the fully unrestricted
5:44open-source version of the open-source
5:47Higgsfield. So, so you won't want to go
5:49to this hosted version. You'll want to
5:52download one of these. So, there's
5:54several ways you can do this. For one,
5:55you can just hit this DMG file, which
5:57literally just downloads it like an app.
5:59Like literally just click the button. I
6:01already downloaded it. I don't want to
6:02download it again. It is a decent amount
6:03of memory, but you just hit this button
6:05right here and you just base it on which
6:07type of computer you have. So, I have a
6:09M3 chip, so you just hit this and it'll
6:11download it. Or you have Mac OS,
6:12Windows, um Linux. So, all these things,
6:15these are different versions you'll
6:16click on. Or you can also install it
6:18from the terminal. You'll just run these
6:20prompts in the terminal and it downloads
6:22it. Super quick and easy. Or you can do
6:24the way which I did and I went on the
6:26GitHub desktop app and I just cloned the
6:29repository. Um literally, it's super
6:31simple. You just go on the app, paste
6:33the link and say clone it. And then now
6:34I am editing it in Visual Studio Code
6:37cuz there are some bugs that I wanted to
6:38fix myself that I've been fixing for
6:40certain models. And so, I've been
6:41running it myself. I edited the parts I
6:43wanted to fix in Visual Studio Code so I
6:45can make it better. So, I made my own
6:47little branch. So, you can do that as
6:48well. Um it's a little more technical
6:51and I have Claude Code helping it. So, I
6:52don't know actually crap. I'm not very
6:54technical at all, but Claude Code knows
6:56it and I just tell it like, "Hey, fix
6:57this part. Fix the loading part. Hey,
6:59the open source model's not working
7:00properly. Fix this." So, there's some
7:03things that I did have to fix and I had
7:04to cater it to my computer in order for
7:06it to work properly. So, once I did
7:08that, I'm running it now and it's in
7:10this app Electron right here. And here
7:12is the fully open source version of
7:14that. Okay? So, now there is an option
7:16to run the API key um or you can do the
7:19local models. As you can see, it is a
7:21little bit more laggy, but here are the
7:23models that I'm running right now. So, I
7:24have Z image turbo, which is a image
7:26generator. I've done a downloaded
7:28Dreamshaper and there's a couple other
7:30ones I haven't downloaded yet. When I
7:32When I do with my variations that I did
7:34in Visual Studio Code, it depends on
7:35what computer you have if you want
7:37certain models to work. So, the only
7:38ones that worked on my computer were
7:40these two. So, they're not perfect, but
7:43um they're pretty low-weight models
7:45because my computer's not the highest
7:46end. So, I downloaded these models and I
7:48was able to generate videos, which is
7:50pretty cool. I'll kind of show you what
7:51I generated, but these are the models
7:53I'm running and literally all you do is
7:54you just hit download and this is how
7:56you download them. Some of them don't
7:59work fully and these have to be debugged
8:01and this is just comes with open-source
8:02stuff. It just does. I hate to say it.
8:04So, yeah, image stuff, if you want to go
8:05to your local image generator, hit this
8:07button right here and it'll pick Z image
8:10turbo. You can also change which one you
8:11want and so I have DreamShaper image
8:14turbo. These are all the ones that you
8:16can download on your computer, the
8:18open-source ones, or you can go back to
8:19the API ones and this is charges it to
8:21that same API. But we don't want that,
8:22obviously. We want the free version. So,
8:24we're going to hit this button and we'll
8:25generate images. Now, it does take
8:26anywhere from like maybe like 1 to 3
8:28minutes to generate an image. So, I
8:29pre-generated a bunch of them to show
8:31you kind of what they look like. So,
8:32this is one right here. This is just
8:33like a backyard that I had it generate
8:35super quickly. This is a car in space.
8:38This is some dude in a fake Iron Man
8:40suit. Um this is Elon Musk in space,
8:43which is um it's not bad. And then this
8:45is a ripped frog. This one actually
8:47didn't work. Um and then this one is a
8:49cat on Mount Everest. So, these are ones
8:51that you can generate with your
8:53open-source Higgs field version. Super
8:55simple. Hit this button, type in your
8:57prompt, then you hit generate and it'll
8:59take a couple seconds to load. Now, I'm
9:01not going to generate one right now
9:02because it is pretty intensive and since
9:04I'm screen recording and doing all this
9:06at once, my computer has 18 gigs of RAM
9:09and it'll probably start getting laggy.
9:11I just want to be cleaner for the demo.
9:13Then, we have the video section. So, the
9:16video section is a little more heavy
9:18than the image section. So, since like I
9:21told you earlier, the only ones that my
9:23computer can actually run are has all
9:25these options, as you can see. Oh, it's
9:27lagging. Um is these two, the LTX video
9:31and the Cog video X2B. And it can make
9:34like maybe 2 to 5-second videos, but
9:37here's some of the results that that I
9:39got from what we were dealing with. So,
9:41like this is the type of stuff that you
9:42could generate, depending on how good
9:43your computer is. Like this one's This
9:46one's pretty cool, not going to lie.
9:47Dang, look at that. Car scene. We have
9:50an elephant jumping into some water.
9:52This one's low resolution, so it's a
9:54little bit easier to generate. This is
9:55another car. Looks like it's going to
9:57jump off of a cliff.
9:59Oh, there it goes. And this is a Looks
10:03like a knife on a rock. This is a ninja
10:05right here.
10:08Okay. This is a car through space. So,
10:10like all these are things that you can
10:11generate. The problem with the
10:13open-source models is like basically
10:15what it means is you are running this AI
10:17model on your computer. And that's how
10:19it's free. It's using your hardware and
10:20not somebody else's hardware. If you're
10:22generating with tokens, you're borrowing
10:25their hardware to generate your video.
10:27And that's why you have to pay a little
10:29bit for it. So, um I was doing a bunch
10:32of research and trying to figure out
10:33which models you can run on which type
10:36of hardware. And I built this graphic
10:38that helped me.
10:39Um so, these are all the main
10:40open-source models, and they're pretty
10:42good. They can make pretty good results.
10:44They're definitely not like as perfect
10:46as like the VO models or the Sora
10:48models. But, these ones are pretty good
10:50that you can run. And this is kind of
10:52how it breaks out. So, you got 1 2.1, 1
10:542.2, High and Video, High and Video 1.5,
10:57um LTX Video, Cog Video, and Cog Video
11:005B. So, I have an M3 Pro right here. And
11:04so, the two that I'm mainly running are
11:06the LTX Video and the Cog Video X2B. And
11:09so, I could run them. It's not perfect.
11:12Um I haven't tried the 1 ones. Um
11:14they're a little bit bigger, so I was a
11:15little more reluctant. But, depending on
11:17like if you have a basic laptop, I'm I
11:19hate to break it to you, it's just not
11:20going to be Your Your Your computer's
11:22not going to have enough processing
11:23power to be able to run an open-source
11:26model. An Intel Mac, neither, cuz those
11:28are kind of old. M1, you possibly could
11:30do it. My version, I can run it. It's
11:32not perfect, but you can do it. M4 Mac
11:34Pro, if you get a new one, yeah, you
11:36should be able to run these ones pretty
11:38well. And if you have like the top-tier
11:40Macs, you should be able to do most of
11:42them pretty easily. Then we go to the
11:44Windows side. This is kind of like the
11:45GPU you'll need and also the memory. And
11:48one thing that's nice about the Windows
11:50side of things, they're a lot more
11:51optimized for models like this, which is
11:53pretty nice. So, looks like the lowest
11:56end you can run a couple of them, but
11:58you there's a few that you can't. And
12:00yeah, as you see we go up in obviously
12:02price and just performance, this is how
12:05you'll be able to run these things. So,
12:06you'll need like the minimum of this
12:07amount of RAM and this type of processor
12:09if you want to run every single one of
12:11these. The cool part though is if you
12:12have this, you can run all these
12:14completely free and locally, which is
12:16awesome. What you'd basically do is you
12:17would just go in here
12:19and you pick your model, like I pick
12:21mine, and then you just prompt it. And
12:24then from there it's literally the same
12:25thing and you're able to generate all
12:27these same videos with that same model.
12:29Now, I'm I am limited to the lower-end
12:31models because I don't have as powerful
12:33of a computer. And that's just how it
12:34goes with working with ups and
12:36open-source models. You also do have the
12:37lip-sync function, same thing. You have
12:39the cinema studio, this one where you
12:40can like build out frames and use these
12:42to generate cinema cinema quality videos
12:45that you can add into the video function
12:47and the image function.
12:48Unfortunately, the workflows and the
12:49agent functions do not work on the host
12:51the self-hosted version yet, but this is
12:53one thing that you could go into your
12:55Visual Studio Code and edit yourself.
12:57And then you have the MCP 2. This is the
12:59full open-source Higgs Field. It's
13:01definitely not a perfect solution if you
13:03have a lower-end computer, but if you
13:05have a decently powerful computer, you
13:08can run all these things without a
13:09problem. You can run the higher-end
13:11models like I showed you on here. So,
13:13yeah, if you have a higher-end computer,
13:15like one with this type of processing
13:17power, you're you're able to run these
13:20higher-end video models, but if you have
13:22somewhere on the lower end, it's just
13:24it's just not viable. Your computer
13:25simply just cannot handle a model like
13:28this. and I and I hate to say it but
13:29it's just the truth of the matter. And
13:31so if you were in this area down here,
13:35I would probably there's lots of like
13:38technically paid softwares that have
13:41free video credits. So you can do maybe
13:44one or two videos free every single day.
13:46I know Google does it, Gemini chat does
13:48it, Grok does it. There's lots of models
13:50that you can create free video. You just
13:52can't do unlimited and there's certain
13:55limitations. But that would be more than
13:57trying to use an open source model
13:58because you just you simply can't. It's
14:01just your computer is just not powerful
14:03enough. And so that's just how it works
14:05with open source models and this is the
14:06reality of the situation. But if you
14:08want to try out this open source model
14:10for yourself, you have those the ability
14:11to do it. By all means go and do it and
14:13it'll be completely free. But yeah,
14:15anyways, this is the open source Higgs
14:17Field. If you have questions leave them
14:18in the comments, but um yeah.