Quote:
Originally Posted by celandina
Sad but true... I still hope that somebody will break the 6 sec. maximum ( to at least 20 seconds) and will figure out how to " uncensor" the adult stuff   |
I'm confused as to why that's so difficult in your case. Just write a custom Python wrapper, load the model weights, LORAs, CLIP models, etc, and direct a multi-stage workflow. Set your progressive weights when you change the LORAs or attention settings (or massively change the prompt step-to-step), and generally stick with Euler and an RTX5090 for "most" workloads.
There isn't some black magic voodoo to it. 30 seconds is no sweat, and there is no "censorship" with any reasonable model.