Hi,
I usually AI locally, but whenever I opened huggingFace to download some model, I found this SCRIPT example including words such as “diffusers” and “Pipe” etc.
After a year or plus, I finally decided to try to understand it,
From my understanding it is a service offered by HF, HG will “host” some AI models and other operations, and you simply need to “call” it by using those lines of code I keep seeing in every HG space.
Question:
I believe this does not cost anything for the user? You just need to log in and sometimes use a token (I used diffusers in the past without even realizing I was just copying some tutorials, but today I am understanding)
The question is, can I use that for a personal project, how much does it cost if I want to scale it up?
I tried to go look into the “pricing” on the HF website, but I am confused.
Can someone tell me more about pricing and using these diffusers?
Example: image generation with Stable Diffusion. (Although I read that runwayml removed the stable diffusion models, mayeb there are other sources anyway?)
Anyhow I would like to get more insight because I did not get the pricing page.
Thanks
Diffusers is just free software made by HF. You can call it a platform.
In other words, it’s on the same level as WebUI, WebUI Forge, and ComfyUI.
WebUI is a GUI, but Diffusers is a CLI, and the file format is a little bit unique.
Converting file formats is easy. (Usually. We’re having a little problem with HF right now)
And the price, but you can do a lot for free.
The $10/month Pro subscription is close to unlimited use remotely to some 70B LLMs.
If you want to run your own programs and models using the GPU in complex ways, or if you want to do continuous image generation, you’ll need to consider a pay-as-you-go plan. (Image generation is probably less expensive at other sites.)
I forgot to mention the most important thing.
If you just use someone else’s public space, it’s free, no matter what or how much you use.
If you use too much at once, you will be restricted a bit, but if you wait an hour, it will be lifted, and there is no long term penalty.
The only difference is that you are on the side of those who publish.
If you’re on the making side, almost all of the spaces below are using Diffusers, Running is free, Running on Zero is $10/month for up to 10 at a time per person, and A4 or so is a pay-as-you-go plan. You can do pretty much anything you want, right?
I forgot to write one more important thing.
If you are going to use it for some larger project, Endpoint API is probably more convenient than Spaces, although it costs more money.
Hello John,
Thanks for your answer,
You mentioned the “$10/month Pro subscription” for close to unlimited use and later one you mentioned end point. I think this is what I failed to understand in term of costs etc.
I will list my concern and hopefuly you will able to guide me furthemore.
First of all, I want to be able to communicate with the image gen from a distance, it can be my computer, my phone, or the phone of a friend. So solutions are: the subscription you mentioned, or some endpoint thing (still not sure about how it works), or even some pay as go.
But, I would prefer it if was possible to have tons and tons of generation possible remotely (in best cases). You said “close to unlimited use”, was that for a SINGLE person use? What if I want to hook up all my friends to my subscription, would your description “unlimited” still hold? Or are these subscriptions more like for single users, and the real “unlimited” use is more for what you call end points? (sorry if this does not make sens I did not learna bout end points yet)
I indeed would like to be able modify the and configure the images in some way (before or after the generation or during). For that you recommanded going for a pay as you go plan, ok questions:
Does this plateform (diffusers), allow you to upload your own model? Or is it limited to a certain “set” of models?
Does it allow you to “set” some other alterations for the outputs, such as having a script running in HG servers having the image as input and creating a new output? This script would need some libraries etc. Whether it is an image you want to alter further, or a text prompt you want to alter. I want to know if we are limited by a simple “input → send to HG → HG uses the plateform to do everything → send you back output”, or are we able to add more steps in this so called “diffusers platform”? And if we are able to use our own models instead of the ones already avalaible.
This pas as you go plan, does it make you pay for simply holding a model in the HG cloud? (that is, is it s possible to have your own model in their cloud to begin with).
Ideally I would like to pay the least possible.
You mentioned possibly less espensive options for image gen in other websites, did you mean, just for generation as a user/consumer, or did you mean as a “developer” who wants his model to be used by his friends etc, and he wants to make his costs the minimum, then he should check other options of other websites?
Hope you can offer a new view on HG capabilities and technologies,thx
First, let me explain about this site as there seems to be a bit of confusion. I’m not that familiar with it…
HF (Hugging Face, ) is like a github with super huge storage and powerful GPU-enabled virtual machines.
Uploads are free, no matter how many terabytes. I’ve uploaded lightly over 4TB myself and left it there.
The only time there is a fee is if you want to use a more powerful GPU or if you want to use a model that is too huge and server-intensive, even if it is for a CPU. (like 70B LLM)
Diffusers is a piece of software used on virtual machines, for images. There is also Transformers for languages and something else for audio. These are not directly related to fees.
You mentioned possibly less espensive options for image gen in other websites, did you mean, just for generation as a user/consumer, or did you mean as a “developer”
As a consumer. If you just want to generate a large number of images using existing models of your choice, PixAI or Civitai, for example, and NovelAI, which is a bit more expensive and model specific, are probably easier and cheaper to operate, since HF is not a site whose main purpose is that.
If you are a developer, here is probably the cheapest and most feature-rich. If you are a developer but also a model maker, Civitai is also good for test generation.
You can roughly think of Diffusers as software that is like a CUI/CLI version of ComfyUI. (although the actual order of appearance is the other way around)
There are so many things that Diffusers can do. In fact, if you can write code, you can theoretically apply it to almost anything related to Diffusion, and it has even been adopted inside the services of many other organizations.
I don’t know much about endpoints because I don’t use them that much, but I think I can teach you a lot about Spaces.
Again, please ask questions based on these assumptions.
P.S.
There are generally three types of cloud services available at HF.
Inference Endpoints
It’s basically paid for and can do the most advanced things… I think, but I’ve never consciously used it, so I’m fuzzy.
Serverless Inference API
As far as I understand it, it is like a degraded free version of Endpoint, slightly unstable and unusable for a little while if accessed too continuously.
And recently, it has only been turned on for a very few models.
Spaces
A service that allows you to publish your own virtual machine with GUI. You can also make it private. The weakest CPU space is free no matter how much you create.
If you subscribe to the $10/month subscription, you can use up to 10 virtual machines at the same time with the Zero GPU service, which can use a powerful GPU for a moment and actually has a slightly higher basic performance.
The rest of the more powerful ones are paid for.
See below for specifics on what you can do. It’s faster to take a look.
There are also several independent, specialized services that do not belong to these groups.
) is like a github with super huge storage and powerful GPU-enabled virtual machines.