Quote:
Originally Posted by Mahadeva
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.
|
I am NOT a geek and DO NOT WANT TO BECOME ONE!!! I make MOVIES the old fashion way BUT want to switch to AI. To that end I need an AI creator who can fucking listen... Hooking up 3 x 7 sec. shots with visible glitches and calling it 20 sec. shot DOES NOT cut it, or exiting screen left and entering also screen left in the next shot ..... Your LORAS RTX5090 progressive weights and the rest of a " mumbo jumbo" is as clear to me as this, to put it simply:
So, if you can make sexy 20 seconds clip in a quality which looks natural let me know, and also since I would be paying you, you need to fucking listen. If you cannot and/or want to stay poor like some of them... then wrap it in your custom Python wrapper and
or
.. It is your choice.