Since the part of the original models might have the root back in NovelAI leak and Instagram based models, according to some rumors, I do not recommend you use it, if you are a hater of NAI leak/Instagram based models and their derivatives.
Since this model is an experimental model to see what will happen when merging a SD 1.x based model to SD 2.x, it is very likely that you get a weird result.
You can run this model on colab WebUI.
Rewrite the following line of this notebook following the instructions I posted here.
Note that this file and only this file in the model is released under public domain (CC0 1.0), in order to keep it compatible with the AGPL license of WebUI.
By replacing the file with the same name in WebUI, you can create your own merged model.
No interpolation is NOT a merging operation. It will work, but it will only return the same model as model A.
Weighted sum can easily destroy the original SD 2.x based model. Multiplier 0.1 was enough for it, whereas 0.01 was OK. There should be a border zone somewhere.
Add difference will work relatively fine, but going too far will likely result in a model similar to this. Additionally, there is no guarantee that you can get the style and/or content you expected to the original SD 1.x model you merged to.
sample outputs
アップしているので、気になるならご自身で見てください。プロンプトはメタデータに入っているはずです。
Check it by yourself if you are interested in this model. The prompts should be in the metadata of each image.
Caution: Since the uploader is a Japanese native, in the event of any differences in meaning between the original Japanese version and a translation, the original Japanese version takes precedence.
要約: ほぼCreativeML Open RAIL-M。但しリバタリアン的解釈によって再構成。CreativeML Open RAIL-Mの制限は、同解釈において維持されているものと判断する。
Summary: A CreativeML Open RAIL-M, interpreted and reconstructed under a libertarian manner. The restriction of CreativeML Open RAIL-M is considered to be valid under such interpretation.
主な相違 (differences from the original CreativeML Open RAIL-M license)
The violation of law or regulation will be judged by and only by your conviction per the presumption of innocence (unless you are convicted, it is not enough to claim it is illegal for the Licensor, even if it looks like it).
ex. Fanart in Japan is technically illegal, unlike countries which have fair use, but as long as it is in the moderate range and the copright holder won't sue you, we will practically do not consider it as problematic.
Outputs you generated by the Model or Derivatives of the Model must be distributed under public domain (CC0 1.0), to ensure not only you but anyone can (re)use it freely.
Note that Dream Studio, run by Stability.ai demands the output be CC0 1.0 as well, but still isn't against the original model license.
Derivatives of the Model will always have to include - at minimum - the same use-based restrictions and the same open permissions.
