A lot of people can be weirdly private about this sort of thing, even though a merge, by definition, is based off of other people's work. You could maybe argue it's different when someone is training a model, but merges are 100% a collaborative process.

I wish it was standard and expected for people to post their recipes.

@JamOnOn I respect that and don't mean to put pressure on you! The statement wasn't really about you, but about how things work in general, and I'll walk back what I said a bit anyway. "I'm sharing this but I really made it for myself" would be an excellent reason to keep a recipe private in any possible art culture. What I wish was different is that I feel like we all tend to keep things secret by default, even when we don't have personal reasons for doing so, and often as a part of "personal branding." That's what's actually bad for art. You definitely aren't part of that.

(Also, I think most people tweak their personal merges so much that they couldn't even tell you what was in them if they wanted to. If I didn't use a spreadsheet to keep track of my own merges, I'd have no idea what I was even doing.)

@JamOnOn I've been running comparison tests and I have about 6 or 7 different merges in each category that all outputs effectively the same output and, when I went back to the CivitAI model site, they have all been merges where the recipe is private. In any of the situations in my comparison tests that contain a merge where the recipe is public, you can often clearly see the influences of the original models, along with the specific personal difference the person merging the models added. And I'm totally for keeping a merge recipe private and I'm definitely not saying that my opinion is law here. It's just that when everyone's "exclusive personal tastes" end up being the effectively the same tastes, CivitAI has to pay more for bandwidth and storage, users have difficulty finding the models that they want to use, and innovation on models is stifled. Quick reminder that closed source development didn't get us as far as we got in the last 4 months as open source development did.

I've run comparisons on about 200+ models at this point with varying parameters and varying prompts. I haven't found a standardized test that I'm comfortable with that I feel covers everything everyone might care about in a comprehensive manner but I have seen a lot of the same seed, same prompt, different models, same/same-ish output.

Anywho, I'll be adding your new version to my tests and I hope I enjoy it!

Probably your personality.