Difficulty understanding the results of the node and how to get good results

[adbrasi]

Hello, first of all thank you very much for your efforts in all this, you are an amazing developer.

I have tested all possible settings and would like to find a way to a better result.

I put in the parameters I liked best and made a comparison.
Parameters:

some information about what I did:
stop_blending_at .80 to make it more creative (I know that blending max is also good for this, but I prefer to use stop_blending).
I've tested it at 20, 40, 60 and 80 steps
Rewind doesn't make much of a change for me, apart from making the colors more vibrant and one or two small details, should I change anything? I don't think it's worth the time it takes.

comparison:

*I uploaded a new image because the other one had an error where the output of the hires fix from the base image was missing the kohya_deep_shrink.

In case it's hard to read, the first image is the base image. The image next to it is the 2 pass txt2img (common hires fix), the one below the base image is the output of the iterative mixer and the last one is the 2 pass txt2img made on top of the output of the iterative mixer.

The iterative mixer output greatly reduces detail. I wonder if it should be like this (because in your examples, the output of the iterative mixer is the final result), so I always add a 20-step ksampler with .5 to .6 denoising to add the missing details, this is the workflow I use and the results are great but as I said, my big question is whether the iterative mixer should give better results, that is, am I missing something?

My system is all up to date.
cfg at 8 for all samplers.
upscale is done with apisr in this comparison but I usually use NNlatentupscale.
I'm using kohya_deep_shrink

Should I use a controlnet and change any parameters?
Any ideas? Am I doing something wrong?

Making it clear: iterative mixer is very good and useful, the result I have with 2 pass txt2img made on top of the output of the iterative mixer is great, but I really wondered if I'm doing something wrong, because in your workflow the iterative mixer is the last process.

result with rewind: