End-user experience?
I'm wondering what the intended end-user experience is for applications running on the Stable Diffusion model.
Quoting from this "Creative Open RAIL-M":
Section III: CONDITIONS OF USAGE, DISTRIBUTION AND REDISTRIBUTION
- Distribution and Redistribution. You may […] reproduce and distribute copies of the Model or Derivatives of the Model thereof in any medium, with or without modifications, provided that You meet the following conditions: Use-based restrictions as referenced in paragraph 5 MUST be included as an enforceable provision by You in any type of legal agreement (e.g. a license) governing the use and/or distribution of the Model or Derivatives of the Model
[…]- Use-based restrictions. The restrictions set forth […] You shall require all of Your users who use the Model or a Derivative of the Model to comply with the terms of this paragraph (paragraph 5).
I Am Not A Lawyer, but my understanding is that use-based restrictions are not GPL-compatible. It explicitly prevents me from adding any such "enforceable provision."
Krita and GIMP are both licensed under the GPL v3, so I can't make any product that both links against either of their libraries and also redistributes this model.
To comply with the letter of the law, I can make an application that works with a user-provided model, and thus offload the burden of responsible redistribution to someone else. But I fear that's the sort of compromise that, while technically viable, leaves no-one satisfied.
It would be disingenuous on my part to pretend that people would use the application for any other purpose than to work with the latest Stable Diffusion model. Most of them don't have the capability to train a model at all, let alone the $0.001 billion dollars necessary to develop something comparable to Stable Diffusion. Of course they are all expecting to use the same model everyone else has been showing off.
I can define the workflow so they have to get a Hugging Face API Token and click through the license agreement on the website before that token grants them access to the model, but nobody wants to sign up for yet another account and give their email address to yet another third party.
That's the sort of experience that's going to either (a) bounce people away before they even get started, or (b) be tolerated, but taints their initial experience with frustration instead of exploration and delight.
(It can be frustrating enough trying to coax the result you want out of an unfamiliar program, we need to save that frustration budget for later!)
I am not unsympathetic to the reasoning behind the use-based restrictions. But how is this supposed to work in practice?