controlnet-sd-2-1-base-v1

<!-- This model card has been generated automatically according to the information the training script had access to. You should probably proofread and complete it, then remove this comment. -->

controlnet-liuch37/controlnet-sd-2-1-base-v1

These are controlnet weights trained on stabilityai/stable-diffusion-2-1-base with new type of conditioning.

Intended uses & limitations

How to use

from PIL import Image
from diffusers import (
    ControlNetModel,
    StableDiffusionControlNetPipeline,
    UniPCMultistepScheduler,
)

checkpoint = "liuch37/controlnet-sd-2-1-base-v1"

prompt = "YOUR_FAVORITE_PROMPT"

control_image = Image.open("YOUR_SEMANTIC_IMAGE")

controlnet = ControlNetModel.from_pretrained(checkpoint, torch_dtype=torch.float32)
pipe = StableDiffusionControlNetPipeline.from_pretrained(
    "stabilityai/stable-diffusion-2-1-base", controlnet=controlnet, torch_dtype=torch.float32
)

pipe.scheduler = UniPCMultistepScheduler.from_config(pipe.scheduler.config)

generator = torch.manual_seed(0)
image = pipe(prompt, num_inference_steps=30, generator=generator, image=control_image).images[0]

image.save("YOUR_OUTPUT_IMAGE")

Limitations and bias

[TODO: provide examples of latent issues and potential remediations]

Training details

Train the ControlNet with semantic maps as the condition. Cityscapes training set is used for training (https://huggingface.co/datasets/liuch37/controlnet-cityscapes). Only 2 epochs are trained for the current version.