A Possibel Update for the Text Encoder.
So TLDR. I finally began to properly use ControlNet (Openpose in this case). And it worked for the most part. Except for the rather major problem that it makes the character a lot smaller.
So, after 6 hours of figuratively (and almost literally) banging my head against a brick wall.
I managed to use several Reference Latent nodes to brute force the correct scale.
From what I read it happens because the node in ComfyUI can automatically resize input images to 1MP (1024x1024) which given I was generating a (1328x2048) Image seems about right.
NO Reference Latent:
With Reference Latent:
Is there a way you could create a version of the node that wouldn't cause this?
When I was testing the Reference Latent Nodes, I got some results that made me prefer, if I didn't have that variable to deal with.
Thanks for your time.