Open Latentnaut opened 8 months ago
It's basically a way to do High Resolution steps in one node. You scan scale up or down steps, and you can also use secondary conditioning if you want (so say take a 2D anime image and then go 3D with second conditioning), and a secondary model that starts on X step you set. You can hook up a upscale model to do the upscaling instead of internally with resampling/latent scaling, or a use a 1x model in the other slot which acts a s a prepass before upscaling. So you could use a 1x denoise model or something.
What are you trying to figure out specifically?
I am in same boat as the OP...
Do you have any visual workflow examples of your options with this node?
Thank you so much
It's basically a way to do High Resolution steps in one node. You scan scale up or down steps, and you can also use secondary conditioning if you want (so say take a 2D anime image and then go 3D with second conditioning), and a secondary model that starts on X step you set. You can hook up a upscale model to do the upscaling instead of internally with resampling/latent scaling, or a use a 1x model in the other slot which acts a s a prepass before upscaling. So you could use a 1x denoise model or something.
What are you trying to figure out specifically?
In my case it's the use of the secondary_model. The changes are very weak even after changing the value of the "secondary_start_cyle" from the default 2 to something else. How do I control the "strength" of the second model?
Hello,
I am looking for any resource that help me to understand Ksampler Cycle better.
Any piece of information is more than welcomed :)