Spaces:
Configuration error
Update README.md
Update README.md to reflect current converted model types. Specifically, adds references to ControlNet capable models, to be identified with a _cn suffix.
Also provides a link to a repo with replacement VAEEncoder.mlmodelc files that enable Image2Image when using models that were converted prior to the use of ml-stable-diffusion 0.4.0 or later in the conversion pipeline.
Does NOT include any reference to obtaining individual ControlNet models, such as Scribble or Canny, however this information could be added to this revision as well.
Should we move the VAEEncoder models under the coreml org? Might be better perhaps for collaboration and centralizing the files.
I would prefer that. At one time, I intended to make a small change in the README.md file at one of my repos and I inadvertently deleted the entire repo, and had to re-upload everything. Things are safer under someone else's control.
My VAEEncoder repo could maybe just be cloned into coreml org as a new model repo? That would be simpler than my re-uploading the pieces to a new coreml org model repo?
Would you also consider cloning my ControlNet individual models (and atatakun's) into coreml org as a single model repo? (The 6 base models would go as regular model repos at coreml org after the landing page is updated per an edit to the edited file in this PR.) Then everything is in one place. I'd be happy to do whatever clean-up is needed in the cloned repos over a day or two. Maybe there would be a way to pin the VAE and CN repos at the top of the model list so they are easy to find?
Afaik there doesn't seem to be a way to "pin" repo's to an org. However we can mention it in the README here and in Mochi Diffusion for easier access.
I'm not sure what
@atatakun
has in his repo but if it makes sense to combine it into one here, go for it 👍
atatakun converted the 14 current ControlNet models (Scribble, Canny, etc.) for Split_Einsum. Mine are 4 different resolutions of the set of 14 for Original. I will leave him a message on Discord tomorrow to ask his permission to clone his repo.
Can you do the actual repo cloning? I'm not sure now to initiate it at coreml org. Or if you can briefly explain to do it, I'm happy to try. Either way, I'll do clean-up of anything that needs it that has been moved. The README.md files, for instance, can probably be simplified and have some links removed.
I will move the CN-ready base models manually (from the original or cloned repo) over to individual model repos at coreml org over the next few days. They each need a README.md page created, and my file names need to be cleaned up to match coreml mode repo standards.
I will keep my repos alive because they are linked in posts here and there, but I will stop making any changes or additions to them after they are cloned to coreml org.
Poking around Hugging Face I found this:
Rename or transfer this model
All links to this model will automatically redirect to the new location, including git operations.
The transfer needs to be done by me, from each of my repos, with the destination set to a new repo name that will be created at coreml. I think as a member of coreml, I have the necessary permissions already.
I will take a deep breath and try to transfer one tomorrow. If it goes well, then the rest. I have a local clone of each repo on my laptop and can upload things manually to coreml org if the transfer messes up.
I will update the README.md in this PR after I get things moved, so I can reference the actual new repo names with links.
The VAEEncoder reo and all of my ControlNet model repo is now in coreml org. One click each and whoosh! Easy as could be. I made minor edits to their README.md pages.
I still ned to get the base models for ControlNet moved over to individual model repos in coreml org. I will do that one a day, starting soon. For now they are in the relocated ControlNet repo at coreml org.
I will update the README.md file in this PR in the next day or two, after I get the first base model moved. I'm going to leave any edits of the README.md at the Mochi git repo to you. You should be able to copy pieces from the next edit I do to the file in this PR.
I'm really glad we are doing this. It will make it much easier for users to find things. And the ControlNet README.md can get fleshed out with instructions for using ControlNet. We need that too, somewhere, and everyone will have to visit that page on the way to getting CN models, so it is a logical spot. The git landing page can just link to all that stuff.
And links to the old repo locations redirect automagically to the new repos at coreml org. Very well done on Hugging Face's part.
I'm moving the last of the CN-compatible base models into its own model repo at Core ML Models. I will refresh my update to the README.md for Core ML Models and try to delete this PR and submit it fresh tomorrow. Everything is in its new place.
Replaced by PR #2