Big Idea

The argument in one line.

Higgsfield's paid AI creation suite has an open-source clone you can self-host for free, trading subscription cost for your own hardware's compute power and some rough edges.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…

You're paying for Higgsfield or a similar all-in-one AI image/video generator and want a free or cheaper alternative.
You have a reasonably powerful GPU (Windows) or Apple Silicon Mac and are willing to run local AI models yourself.
You're comfortable installing a desktop app or cloning a GitHub repo, and troubleshooting the rough edges of an open-source project.
You want to understand the tradeoff between hosted pay-per-token AI tools and self-hosted free-but-BYO-hardware tools.

SKIP IF…

You have a basic laptop or an older Intel Mac — the video is explicit that these can't run the local models at all.
You want a polished, bug-free product; the creator repeatedly notes broken features and models that don't work on his hardware.
You're not willing to touch a terminal, GitHub, or a code editor to fix bugs yourself.

TL;DR

The full version, fast.

Open Generative AI is a free, open-source project on GitHub that rebuilds Higgsfield's entire interface — image studio, video studio, lip sync, cinema studio, marketing studio, workflows, and agents. There are three tiers: the expensive official Higgsfield, a hosted version of the open-source clone that charges cheaper pay-per-use API tokens across many models, and a fully local version you download and run on your own computer for free. The free tier works by using your own GPU or Apple Silicon instead of renting server compute, so it costs nothing but is limited by your hardware — the creator's M3 Pro could only run two lightweight video models (LTX Video and CogVideoX) reasonably well, and he includes a compatibility chart showing which models need which GPU/RAM tiers. The practical conclusion: the free local route only works if your machine is powerful enough, and even then some features (workflows, agents) aren't ported to the self-hosted version yet.

Free for members

Chat with this breakdown — free.

Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.

Create a free account →

Chapters

Where the time goes.

00:00 – 00:23

01 · Cold open + what this project is

Introduces Open Generative AI, an open-source clone of Higgsfield built by another creator and hosted publicly on GitHub for anyone to download and edit.

00:23 – 00:57

02 · Two ways to use it

Explains the fork between the hosted pay-per-use version and the fully local free version, promising to cover both.

00:57 – 05:00

03 · The hosted paid tier walkthrough

Tours the hosted version's Image Studio, Video Studio, Lip Sync, Cinema Studio, Marketing Studio, Workflows, Agents, and Apps sections, explaining the API-key billing model and how it undercuts Higgsfield's own pricing and other model vendors.

05:00 – 06:13

04 · Pricing tier chart + intro to self-hosting

Shows a 3-tier price/difficulty chart (Higgsfield > Hosted Open Source > Local Open Source) and transitions to the GitHub repo for the fully local, free version.

06:13 – 07:18

05 · Installing the desktop app

Walks through installation options — one-click DMG/installer, terminal command, or cloning via GitHub Desktop — and mentions using Claude Code to fix bugs and adapt the code to his own machine.

07:18 – 09:17

06 · Running local models — image + video

Demos the Electron desktop app, selecting local vs API models, downloading Z Image Turbo and Dreamshaper for image generation, and shows pre-generated sample images and short video clips from LTX Video and CogVideoX.

09:17 – 11:05

07 · Hardware compatibility reality check

Explains that local generation uses your own GPU instead of rented compute (hence free), and unveils a compatibility chart mapping open-source video models against Mac and Windows hardware tiers.

11:05 – 12:57

08 · Using the local generator + remaining gaps

Shows picking a local model and generating, notes workflows/agents don't work in the self-hosted build yet, and reiterates the hardware ceiling for higher-end models.

12:57 – 14:21

09 · Fallback advice + close

For underpowered machines, suggests stacking daily free-credit allowances from other paid AI tools (Google, Grok, etc.) instead of trying to self-host, then signs off.

Atomic Insights

Lines worth screenshotting.

Open Generative AI is an open-source clone of Higgsfield's entire product surface — image, video, lip sync, cinema studio, marketing studio, workflows, and agents.
The project has three cost tiers: official Higgsfield (most expensive), a hosted clone billed per-token (cheaper), and a fully local self-hosted version (free but uses your own hardware).
Running AI models locally is free because you're using your own GPU/CPU instead of paying to rent someone else's server hardware.
A creator's own hardware directly caps which models are usable — on an M3 Pro Mac, only two of roughly seven video models (LTX Video and CogVideoX) ran acceptably.
Older Intel Macs cannot run these open-source video models at all, regardless of willingness.
Windows GPUs tend to be more optimized for local model inference than comparable Mac hardware, per the creator's compatibility research.
The hosted middle-tier still requires generating an API key and loading funds, but rates are cheaper than paying each underlying model vendor (e.g. Midjourney) directly.
Workflows and agent features exist on the hosted version but are not yet ported to the self-hosted local build.
For underpowered machines, the more realistic free path is stacking daily free-credit allowances across multiple paid AI tools (e.g. Google, Grok) rather than trying to self-host.
The creator used Claude Code to help debug and modify the open-source codebase for his own machine despite describing himself as non-technical.

Takeaway

Self-hosting AI tools trades subscription cost for hardware and patience.

WHAT TO LEARN

An open-source clone of a paid AI suite can run entirely free on your own machine, but the free tier is gated by GPU power, not willingness, and ships with real rough edges.

Open-source clones of popular paid AI products often exist and replicate the full feature set, not just a subset — check GitHub before assuming a tool only comes as a subscription.
Local/self-hosted AI generation is free because it spends your own hardware's compute instead of renting a server's — the tradeoff is speed and model selection, not just setup effort.
Match the model to your hardware before assuming self-hosting will work: a compatibility chart (GPU/RAM tier vs model) is worth building or finding before you sink time into installation.
A hosted-but-cheaper middle tier (pay-per-token against an open-source backend) can undercut both the original paid product and going direct to individual model vendors.
Open-source projects will have broken or unported features (here: workflows and agents don't work in the self-hosted build) — budget time to work around gaps, not just install and go.
When self-hosting isn't viable on your hardware, stacking free daily credit allowances across multiple existing paid AI tools is a more realistic free path than fighting your GPU's limits.
Non-technical users can still modify open-source codebases for their own hardware by pairing a coding assistant (here, Claude Code) with a willingness to say 'fix this part' rather than reading the code themselves.

Glossary

Terms worth knowing.

Open Generative AI: The open-source GitHub project (by a creator using the handle anil-matcha) that recreates Higgsfield's AI content-generation interface for free self-hosted use.
Local/self-hosted model: An AI model that runs on your own computer's hardware instead of a remote server, so there's no usage fee but performance depends entirely on your GPU/RAM.
LTX Video: A lightweight open-source video generation model capable of running on modest consumer hardware, one of the few models the creator could run locally.
CogVideoX: An open-source video generation model (used here in its 2B variant) that the creator ran locally alongside LTX Video for short video clips.
Cinema Studio: A feature in the Open Generative AI / Higgsfield interface for building frame-by-frame cinematic sequences rather than single generations.

Resources

Things they pointed at.

00:15linkOpen Generative AI (GitHub) ↗

00:00productHiggsfield AI ↗

06:11toolClaude Code

Quotables

Lines you could clip.

10:10

“The problem with the open source models is, like, basically what it means is you are running this AI model on your computer, and that's how it's free. It's using your hardware and not somebody else's hardware.”

crisply states the entire rent-vs-own tradeoff in one breath→ TikTok hook↗ Tweet quote

13:50

“Your computer simply just cannot handle a model like this. And I hate to say it, but it's just the truth of the matter.”

honest, unhyped disclaimer that cuts against typical AI-tool hype→ newsletter pull-quote↗ Tweet quote

The Script

Word for word.

Read-along

Don't just watch it. Burn it in.

See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.

00:00Okay. So today I'm gonna show you guys how to get the open source Higgs field. There are several ways you can do it, but this is kind of how it looks when you get it going.

00:07This is the whole system right here. It's pretty cool what you can do. Okay.

00:11So this is where it all comes from. It comes from a Neil Macha. The jet it's called the project is called open dash generative dash a I and you have all the files right here.

00:20So if you want, you can download it yourself and you can edit it however you want. There are some parts that don't work perfectly and that's because it's an open source project.

00:29That's kinda how this works. It's like generated by some person and then made for free and they're giving this out for you to use and try. Okay.

00:35So there's two ways that you can use this open source Higgs field. So we have the pay per use version and the free version. I'll walk through both of them, and I'll give you the benefits and the drawbacks.

00:46Let's do this. Okay. So first, I'm gonna show you the paid version.

00:48So basically, the paid version is super easy to get. All you have to do is you go to the hosted version, which is this website right here on my screen. You click on that and it brings you straight to this thing that I was showing you earlier.

01:02Then from there, this allows you to use every single model in existence I can think of. I I'm sure there's more models out there, but there's just a ton on here.

01:11And there's lots of options and it runs really well because it's running on their server. But basically what you do for this one is it is the paid version, but it is a lot cheaper than using something like Higgs Field. Like this is the open source version, so this is kinda like you have like the tiers of Higgs Field AI is very expensive, then you have middle, which is kind of middle of the road, which is this one.

01:31Then you have the fully open source where you download on your own computer, and that is free, basically. So I'll show you how the middle tier works, and then we'll get into the free one. Okay.

01:39Middle tier, what you're gonna wanna do is you will have to generate an API key. This is a Muapai. I don't know how to pronounce this properly.

01:45Basically, you're gonna do is generate an API key in here, and it's like a mega API key. And so it covers every single model you can think in existence.

01:54A lot of them are cheaper than doing it through the actual company, so that's nice. But you have access to tons and tons of models. So if you get really good with a certain model that's super cheap, you can generate a ton of stuff for a lot cheaper than it would be if you were to generate it like with Midjourney or or other companies that host it on their servers.

02:13So it's a lot cheaper in that way. So it's not completely free but it is a lot cheaper which is pretty nice. But basically you have all these options and how it works is it connects with this API key.

02:23You just generate the API key. I I loaded $10 on there and I've been generating a ton of stuff. And that's when you have this section right here.

02:30So you're able to basically generate anything through Nano Banana, Flex Dev, and there's tons of options too. You can generate four at a time. You can change the aspect ratio.

02:38You can add reference images. There's just a ton of things that you can do. That's for the image stuff.

02:43Same thing with the video. You have all the options. You have like a bajillion options, and you can even download do the open source models.

02:50They do host them on their own server. But the thing I'll show you later is you can host these models also on your own computer and generate these videos for free. I'll get to that.

02:58I'll get to that. There also are several models that you cannot host on your own computer, and that's like Seedance and Cling and Vio.

03:05These are all hosted on big fat servers because they take a lot to generate video. But you're able to do all those at a more affordable price here in the video section.

03:16So you can change the aspect ratio, change the time, you can change the quality, add other images, like there's a lot of stuff you can do. You also have this lip sync section where you can make it so it matches up with the voice of a face and you can throw in audio. Pretty simple.

03:29Then you have cinema studio. This is where you can generate frame by frame and kinda like build the stuff together.

03:35And then you have marketing studio which is where you can build out ads for it. This is also much more affordable because you can choose the model and you can choose the variant, how you want it to look, the avatar. Like there's a ton of cool options in here.

03:46Then also here you do have workflows where you can do certain things which are like more specialized. So you do social media, ecommerce, fashion, featured, featured, social media, like all these things are are things that can help you like build better things and you can build your own workflows as well and community stuff which is generated from the community, which is cool.

04:04I like that it's all kind of centralized into one system. It's like basically the community building their own version of Hicksfield that's kind of all combined here. So you don't have to pay for Hicksfield prices.

04:13So you just pay like the base price and then everybody contributes to everybody. It's it's a pretty cool little system. It's not perfect though.

04:19It's not perfect. I'm not gonna act like this is completely free because it's not. There is a free way to do it.

04:24It's not this one exactly. But this is the open source version. So like I guess in the sense of open source, like anybody can contribute to this platform right here and help build it to make it better, which is pretty cool.

04:34You also do have agents right here where you can literally just like like, look at this. This is crazy. Like a veggie talk creator.

04:41Like, there's a ton of different agents they can use to make content, which is pretty cool. All of these, you will have to pay for the the tokens, but and then you also have apps where you can literally just build out your own template and create your own, I guess, wrapper in the sense of like you can create your own AI video studio and sell that to other people, which is pretty cool.

05:00And this one you can just like straight up go to the GitHub and do it from there and build it out. And this is completely free. You would have to pay for the tokens, but if you're charging somebody like twice as much for the tokens, then it can make sense and it can be an actual business, which is pretty cool.

05:11All these GitHubs are right there and you can grab the templates and just download them and create your own AI business just like that. It's pretty gnarly how quick and easy it is. That is the main stuff on the hosted version.

05:23Alright? Now I know you guys didn't come here for the hosted version. I just wanna show that this is kind of what the creator is making.

05:29It's like a cheaper and open source alternative to doing Higgs Field. But there is like the other part where you can just host it on your own computer. So let me show you kind of what that looks like.

05:38So you're gonna go into here, right, go back to the main GitHub where I was, and this is where you can get the fully unrestricted open source version of the open source Higgs field. So so you won't wanna go to this hosted version.

05:51You'll want to download one of these. So there are several ways you can do this.

05:55For one, could just hit this d m g file which literally just downloads it like an app. Like literally just click the button. I already downloaded it.

06:02I don't wanna download it again. It is a decent amount of memory, but you just hit this button right here and you just base it on which type of computer you have. So I have a m three chip, so you just hit this and it'll download it.

06:11Or your mac o s, windows, linux, so all of these things, these are different versions you'll click on. Or you can also install it from the terminal.

06:20You'll just run these prompts in the terminal and it downloads it. Super quick and easy. Or you can do the way which I did, and I went on the GitHub desktop app and I just cloned the repository.

06:30Literally, it's super simple. You just go onto the app, paste the link and say clone it. And then now I am editing it in Visual Studio Code because there are some bugs that I wanted to fix myself that I've been fixing for certain models.

06:41And so I've been running it myself. I edited the parts I wanted to fix in Visual Studio Code so I can make it better. So I made my own little branch.

06:48So you can do that as well. It's a little more technical and I have Claude Code helping you. So I don't know actually crap.

06:54I'm not very technical at all. But Claude Code knows it and I just tell it like, fix this part. Fix the loading part.

06:59Hey, the open source model's not working properly. Fix this. So there's some things that I did have to fix and I had to cater it to my computer in order for it to work properly.

07:07So once I did that, I'm running it now and it's in this app, Electron right here. And here is the fully open source version of that. Okay?

07:15So now there is an option to run the API key or you can do the local models. As you can see, is a little more laggy but here are the models that I'm running right now.

07:24So I have z image turbo, which is a image generator. I've download downloaded Dreamshaper and there's a couple other ones I haven't downloaded yet. What I what I did with my variations that I did in Visual Studio Code, it depends on what computer you have if you want certain models to work.

07:38So the only ones that worked on my computer were these two. So they're not perfect but they're pretty low weight models because my computer's not the highest end.

07:47So I downloaded these models and I was able to generate videos which is pretty cool. I'll kinda show you what I generated but these are the models I'm running and literally all you do is you just hit download and this is how you download them.

07:58Some of them don't work fully and these have to be debugged. And this just comes with the open source stuff. It just does.

08:03I hate to say it. So yeah, image stuff. If you want to go to your local image generator, hit this button right here and it'll pick z image turbo.

08:10You can also change which one you want. And so I have Dreamshaper image turbo. These are all the ones that you can download on your computer, the open source ones.

08:19Or you can go back to the API ones and this charges it to that same API. But we don't want that obviously, we want the free version. So we're gonna hit this button and we'll generate images.

08:26Now it does take anywhere from like maybe like one to three minutes to generate an image. So I pre generated a bunch of them to show you kinda what they look like. So this is one right here.

08:32This is just like a backyard that I had to generate super quickly. This is a car in space. This is some dude in a fake Iron Man suit.

08:41This is Elon Musk in space, which is it's not bad. And then this is a ripped frog. This one actually didn't work.

08:48And then this one is a cat on Mount Everest. So these are ones that you can generate with your open source Higgs field version. Super simple.

08:56Hit this button, type in your prompt, then you hit generate, and it'll take a couple seconds to load. Now I'm not gonna generate one right now because it is pretty intensive. And since I'm screen recording and doing all this at once, my computer has 18 gigs of RAM and it'll probably start getting laggy.

09:12I just want it to be cleaner for the demo. Then we have the video section. So the video section is a little more heavy than the image section.

09:20So since, like I told you earlier, the only ones that my computer can actually run are it has all these options as you can see. Oh, it's lagging. Is these two.

09:29The LTX video and the COG video x two b. And it can make like maybe two to five second videos. But here's some of the results that that I got from what we were dealing with.

09:41So this is the type of stuff that you could generate depending on how good your computer is. Like this one's this one's pretty cool. Not gonna lie.

09:47Dang. Look at that. Car scene.

09:49We have an elephant jumping into some water. This one's lower resolution so it's a little bit easier to generate.

09:55This is another car. It looks like it's gonna jump off of a cliff. Oh, there it goes.

10:01And this is a looks like a knife on a rock. This is a ninja right here.

10:08Okay. This is a car through space. So, like, all these are things that you can generate.

10:12The problem with the open source models is, like, basically what it means is you are running this AI model on your computer, and that's how it's free. It's using your hardware and not somebody else's hardware. If you're generating it with tokens, you're borrowing their hardware to generate your video, and that's why you have to pay a little bit for it.

10:30So I was doing a bunch of research and trying to figure out which models you can run on which type of hardware and I built this graphic that helped me. So these are all the main open source models and they're pretty good.

10:43They can make pretty good results. They're definitely not like as perfect as like the VIO models or the SOAR models, but these ones are pretty good that you can run.

10:51And this is kind of how it breaks down. So you got one two point one, one two point two, high end video, high end video 1.5, LTX video, COG video, and COG video five b.

11:01So I have an m three pro right here. And so the two that I'm mainly running are the LTX video and the COG video x two b. And so I could run them.

11:10It's not perfect. I haven't tried the WAN ones. They're a little bit bigger so I was a little more reluctant.

11:16But depending on like if you have a basic laptop, hate to break it to you, it's just not gonna be your computer's not gonna have enough processing power to be able to run an open source model.

11:26An Intel Mac neither because those are kind of old. M one, you possibly could do it. My version, can run it.

11:32It's not perfect but you can do it. M four Mac Pro, if you get a new one, yeah, you should be able to run these ones pretty well. And if you have like the top tier Macs, you should be able to do most of them pretty easily.

11:43Then we go to the Windows side. This is kind of like the GPU you'll need and also the memory. And one thing that's nice about the Windows side of things, they're a lot more optimized for models like this which is pretty nice.

11:54So looks like the lowest end you can run a couple of them, but there's a few that you can't. And yeah, as you see we go up in obviously price and just performance. This is how you'll be able to run these things.

12:06So you'll need like the minimum of this amount of RAM and this type of processor if you want to run every single one of these. The cool part though is if you have this, you can run all of these completely free and locally, which is awesome. What you'd basically do is you would just go in here and you pick your model, like I pick mine, and then you just prompt it.

12:24And then from there, it's literally the same thing. And you're able to generate all these same videos with that same model. Now I'm I am limited to the lower end models because I don't have as powerful of a computer.

12:34And that's just how it goes with working with open source models. You also do have the lip sync function, same thing. You have the cinema studio, this one where you can like build out frames and use these to generate cinema quality videos that you can add into the video function and the image function.

12:48Unfortunately, the workflows and the agent functions do not work on the host the self hosted version yet. But this is one thing that you could go into your Visual Studio Code and edit yourself. And then you have the MCP two.

12:59This is the full open source Higgs field. It's definitely not a perfect solution if you have a lower end computer. But if you have a decently powerful computer, you can run all these things without a problem.

13:10You can run the higher end models like I showed you on here. So yeah, if you have a higher end computer like one with this type of processing power, you are you are able to run these higher end video models. But if you have somewhere on the lower end, it's just not viable.

13:25Your computer simply just cannot handle a model like this. And I hate to say it, but it's just the truth of the matter. And so if you were in this area down here, I would probably there's lots of like technically paid softwares that have free video credits.

13:43So you can do maybe one or two videos free every single day. I know Google does it. Gemini or Chat does it.

13:49Grok does it. There's lots of models that you can create free video. You just can't do unlimited and there are certain limitations.

13:56But that'll be more than trying to use an open source model because you just you simply can't. It's just your computer is just not powerful enough. And so that's just how it works with open source models, and this is the reality of the situation.

14:08But if you wanna try out this open source model for yourself, you have those the ability to do it, by all means, go and do it, and it'll be completely free. But yeah. Anyways, this is the open source Higgs field.

14:17If you have questions, leave them in the comments. But yeah.

The Hook

The bait, then the rug-pull.

Higgsfield built a slick paid interface for AI image and video generation — and someone open-sourced a clone of the whole thing. This walkthrough shows both ways to use it for less: a cheaper pay-per-token hosted version, and a fully free version that runs on your own computer instead of a rented server.

Frameworks