Getting to the meat of the release (from their github [1]):
> What's the difference between StableStudio and DreamStudio?
Not much! There are a few tweaks we made to make the project more community-friendly:
- We removed DreamStudio-specific branding.
- All "over-the-wire" API calls have been replaced by a plugin system which allows you to easily swap out the back-end.
- On release, we'll only be providing a plugin for the Stability API, but with a little bit of TypeScript, you can create your own.
- We removed Stability-specific account features such as billing, API key management, etc.
- These features are still available at DreamStudio's account page.
Doesn't look like any of their models are getting open sourced alongside the app but they're making it possible to swap out the image generation backend using plugins
Yes, the OpenRAIL licences are deliberately not open source. Being open source or not is binary, and proprietary is the antithesis of open source.
There are of course additional ways to describe specific proprietary things, like distributable in this case. Distributable but non-free, because the software license is morally judgemental, and limits 'fields of endeavour'.
I'm with you on "not open source", but I don't think "proprietary" works as the opposite of "open source" here.
The dictionary definition of "proprietary" tends towards "one that possesses, owns, or holds exclusive right to something" - that doesn't quite fit here, because models licensed under OpenRAIL aren't exclusively restricted to their owners.
They have terms of how they can used that are more restrictive than open source licenses, but they don't enforce "exclusive" usage by their owners.
Model weights can't really be described as source code though. The equivalence isn't exact, but I'd describe the weights more as the compiled binary, with the training data & schedule being the source (which is sort of under an open source license, with the complication of LAION's "it's just links"). The fact it costs $1 million to "compile" isn't relevant.
This isn't to defend Stability particularly though - they've been getting increasing slow and restrained in their model releases. Charitably because they're attracting a lot of heat from political and anti-AI aligned groups. Uncharitably because they've taken a lot of funding now.
> Model weights can't really be described as source code though. The equivalence isn't exact, but I'd describe the weights more as the compiled binary, with the training data & schedule being the source
I think this is a really interesting discussion! I see where you're coming from, but I'm minded to disagree in part.
For one, I think it's possible to release model weights under a liberal licence, yet train on proprietary data. (ChatGPT is trained on oodles of proprietary data, but that doesn't limit what OpenAI do with the model). Normally, obviously, the binary is a derivative work of the source.
Also, the GPL defines source code as 'the preferred form for modification'. I don't disagree that model weights are a black box. But we've seen loads of fine tuning of LLaMA, so we don't always need to train models from scratch.
Ideally, of course, having both unencumbered training data and model weights would be perfect. But in the interim, given I don't have that million dollars, I'll settle for the latter.
Yeah, neither view is a perfect fit. Another example is vision transformer backbones, where a common generic base weight is used to fine-tune all sorts of different processes (segmentation, image to text, etc). The terminology (and licenses) haven't really kept up.
A properly unencumbered model would be my preference too. The community generally seems a bit laissez-faire with license compliance though, so the restrictions currently don't generate much push back. (Plus it's not totally clear that you can copyright model weights at all, given they're the output of an automatic process).
How is Stability AI going to make money? What they provide is super cool and VCs are into it but at some point they need to start giving them returns. Is the plan to get acquired?
Their business plan is to build custom models for clients/companies who want to own models built on their own data. No comment on the viability of such a business model.
You can buy credits to generate images using DreamStudio. In theory, this seems more convenient than running software locally. But last I checked, their image quality wasn’t all that interesting compared to MidJourney, so I stopped using it.
The Discord-based UI for MidJourney is clunky and unreliable, and waiting for images to generate is annoying. Image quality has improved quite a bit, but it’s no better at following instructions. So it seems like there’s an opening for a competitor, but the competition would need to actually be better.
When even users are recommending a subscription service, that's when you know we have become brainwashed into thinking it's a healthy or required practice.
Here are a few things we’d be excited to support…
Local inference through WebGPU
Local inference through stable-diffusion-webui
Desktop installation
ControlNet tools
Tell us what you want to see!
Looks like they are trying to make it worth with a1111.
Here are a few things we’d be excited to support…
Local inference through WebGPU
Local inference through stable-diffusion-webui
Desktop installation
ControlNet tools
I wouldn't bet on it. Automatic1111 already has a strong community and growing ecosystem. Stability.ai would have to try really hard and offer something significant to win people over.
I’d rather them focus on getting their hosted solution/api on par with Automatic. It needs control net, it needs seamless images. And all the other goodies that are now pretty standard.
I like a hosted solution. Can we just make it much better please? If I am paying you to host a service I don’t want to also have to go an give my labour to create the software. That’s why I wanted a hosted service in the first place.
For serious generation control, nothing else comes remotely close. It is the most impressive github project I have every installed tbh. The flexibility of the system is unmatched in the image gen space. It feels like I am back in the 90s.
So StableStudio is just a plugin-extendable user interface for interacting with an image generation model (e.g. via queries against some opaque remote endpoint) , but does not provide the actual image generation model?
So the actually interesting meat is not part of the offer?
Don't discount the value (or difficulty) of a great user interface
There are multiple servers that provide the backend "meat" already: gyre.ai (my personal project); Automatic1111 webui; InvokeAI; plus all the hosted APIs to name just a few.
There are no UI interfaces I'd consider great. I think Flying Dog aistudio[1] is the best (although I'm biased), but still early days. The UI part of WebUI is ... functional(ish).
DreamStudio / StableStudio is super basic in capability at the moment, but it's a space with tons of room for improvement.
If you want to use stable diffusion t2i and i2i today via an self-hosted api it's very easy to do so with auto1111, just click the api link at the bottom of the gradio app and there's extensive documentation on the endpoints.
I think it can only use other APIs itself. Looks like they want to make it work with a1111's api and they will ship a plugin that talks to stability's pay api.
> What's the difference between StableStudio and DreamStudio?
Doesn't look like any of their models are getting open sourced alongside the app but they're making it possible to swap out the image generation backend using plugins[1] https://github.com/Stability-AI/StableStudio