Sign In

How to Make a good Merge

How to Make a good Merge

How to Male a good Merge

This guide will tell you how to make a good merge by telling you how I've failed at making models in the past.

Section One: Good Base Merge

So of Course AI has to be complicated after all it's a crap load of math which myself and literally anyone who isn't an AI could understand, But just like LLMs Stable diffusion can be merged and something I tried with LLMs works with Stable diffusion.

When merging models to get one of the huge merges I am known for like Furation or MasterAnime, What you have to do is Get a bunch of models with features you like and want to extract, BUT after merging some of the model which are your favorite, keep them around, I suggest usually 6-9 models.

Section Two: Wondering why the merge sucks

Now if you do what I used to do, You'll have a model you kinda don't like which is good, becuase as long as it has Some* of the features you like it'll work. So what you do is use add difference with to merge those models which are your favorite which I mentioned earlier and if you don't have the super merger extension, use weight sum.

Section Three: Making the Merge Good

Now what you do when you have the Base merge and the merge from your favorite models which from now I'll be calling the "Refiner Merge" You will need to test both of them, But if you don't want to you can just use Add difference with Smooth Add to add together the models at the Alpha of 1. IF you want to test both models Choose one of the ones you like better And make it the first model then play around with the Alpha until you are happy.

Section Four: BUT THATS JUST A THEORY, A STABLE DIFFUSION THEORY

Heh Game Theory reference...

Anyways This is just a Theory, after all this worked well for one model and that was Furation BUT if you get good results, I'd like to know.

4

Comments