This was a failed series of automerges that I attempted. Unfortunately, there were massive methodological issues that led to me discontinuing the project. The models used were diverse though, so they perform pretty well.
Description
Marigold and Macaroni.
For some reason, this model mostly inherited marigold's weights, don't ask me why. It do be lookin' good tho. 😊
FAQ
Comments (4)
They are all an interesting series of models, but 4gb each would become too much: any chance to reduce them to smaller 144mb LORAs ?
These are merges, converting them to LoRAs would lose lots of detail and they would be much larger than 144mb.
@Azazelle Yes it would lose some detail but you would retain the vast majority of the data and still be able to upscale and combine with other models. Plus all of the Loras would be roughly 0.5GB (maybe not 144 MB) or less which is much smaller than 4GB. Take the Dalcefo Anime Pastel Mix and its Lora for example. Massive size difference with high quality retained
@Azazelle I've seen mostly 144mb LORAs, some 350mb and a few even more, but all still very much smaller than the full 2gb-4gb-7gb models, with a really good quality. Anyway, to the author the choice.