How was the Verus Vision model trained

Hi everyone! Today I will tell you about how the Verus Vision model was trained.

Now, I will show the parameters I used for training LoRAs on CivitAI.

Where X is a number that should be equal in both parameters, for example, 32 and 32, and so on.

I used 100 images of approximately one concept to train one LoRA.

These may not be the best parameters, but they are the best out of those I tried.

To obtain the final Flux model, I used the method of merging the final LoRA into the model. The final LoRA model was obtained by merging 10 LoRAs with different concepts using the Concat method (which you can find in Kohya_ss in the sd3-flux.1 branch, then in the LoRA tab -> Tools -> Merge FLUX LoRA). Below, I provide the data based on which I merge the models, and the merging was done sequentially.

Now, let's talk about how to merge the Flux model and LoRA. Below is a very simple ComfyUI workflow for merging LoRA into the model.

I think that's all I wanted to tell you. I hope this will be useful for you.