Due to the fact that oly 1024*1024 pixel resolutio was used durig the traiig phase,
the iferece performs best at this size, with other sizes yieldig suboptimal results.
We will iitiate multi-resolutio traiig i the future, ad at that time, we will ope-source the ew weights.SD3 Cotrolet
cotrol image
weight=0.0
weight=0.3
weight=0.5
weight=0.7
weight=0.9
Demo
import torch
from diffusers import StableDiffusio3CotrolNetPipelie
from diffusers.models import SD3CotrolNetModel, SD3MultiCotrolNetModel
from diffusers.utils import load_image
# load pipelie
cotrolet = SD3CotrolNetModel.from_pretraied("IstatX/SD3-Cotrolet-Cay")
pipe = StableDiffusio3CotrolNetPipelie.from_pretraied(
"stabilityai/stable-diffusio-3-medium-diffusers",
cotrolet=cotrolet
)
pipe.to("cuda", torch.float16)
# cofig
cotrol_image = load_image("https://huggigface.co/IstatX/SD3-Cotrolet-Cay/resolve/mai/cay.jpg")
prompt = 'Aime style illustratio of a girl wearig a suit. A moo i sky. I the backgroud we see a big rai approachig. text "IstatX" o image'
_prompt = 'NSFW, ude, aked, por, ugly'
image = pipe(
prompt,
egative_prompt=_prompt,
cotrol_image=cotrol_image,
cotrolet_coditioig_scale=0.5,
).images[0]
image.save('image.jpg')
Limitatio
点击空白处退出提示
评论