Lightricks
STG Guidance shouldn't be mandatory
I am trying to get a workflow without STG but I can't.
For one, there is no explanation anywhere in the github repo or elsewhere as to what it does. I pretty much figured it out, but its still quite annoying, since its a core component and the actual UI is not remotely accessible, in that it gives literally 0 functional overview of what it does.
Secondly, I get what it's doing, but it interferes with another method I use to attenuate the positive conditioning, and I'd very much like a workflow without it.
My main gripe with many workflows is when they restrict use cases as this tends to make the community less able to explore the tool effectively. If i so much as google STG, I get 3 entire results. This makes your tool, right off the bat, entirely inaccessible to the vast majority of people.
I hate stg... it feels like a cop out because they cant train the model properly.
I was able to not use STG until now in my workflows from .95-.96, i used perturbed attention. This new 13b model seems like it needs specific bs settings to even work.
Their upscale does 2x upscale... cant change that? cant do 1.5? 1.2? I tried using my own upscale method, but it blurs details because i cant use that odd upscale model with any other nodes... I need to use another.
I LOVED LTX 0.95. It has done nothing but go backwards since.
