how on earth do you get such accurate style transfers with t2iadapter_style?
I've recently came across these early March tutorials on style transfer (using ControlNet and t2iadapter_style). Somehow, despite following the setup and tutorials, I could never seem to just transfer a style to another image. I'd even used two ControlNets - one for Canny and one for Hed and still nothing. All I get are completely random images that make no sense.
Did something change on ControlNet and T2i since then? Am I missing something? Can somebody who managed to figure it out help put me on the right path?
eg. I've removed --medvram in my bat file. I've renamed sd14v1 to -fp16 and then changed them back. Basically I've scoured the forums and tried to follow different solution advice but nothing sticks. What am I missing here?
Did you add the renamed .yaml file alongside the model? (T2I-Adapter Support part in https://github.com/Mikubill/sd-webui-controlnet)
Also, you don't need to remove –medvram. In my experience, you only need to add --always-batch-cond-uncond.
yep all renamed alongside their equivalent .pth files, with the except of t2iadapter_seg_sd14v1 and the rest of the control_sd15xxx.pth files. all right, i'll read the -medvram and --always-batch commands back in.
figured it out. It was a gradio upgrade that i needed. lol
ok so with controlnet 1.1, clip_vision is no longer there, but t2iadapter_clip_vision is. I can only use a t2iadapter like canny or sketch if I want to do a style transfer with t2iadapter_style right?
ok so with controlnet 1.1, clip_vision is no longer there, but t2iadapter_clip_vision is. I can only use a t2iadapter like canny or sketch if I want to do a style transfer with t2iadapter_style right?
How to perform style conversion in Controlnet 1.1