|
--- |
|
license: other |
|
base_model: stabilityai/stable-diffusion-xl-base-1.0 |
|
tags: |
|
- stable-diffusion-xl |
|
- stable-diffusion-xl-diffusers |
|
- text-to-image |
|
- diffusers |
|
- controlnet |
|
inference: false |
|
--- |
|
|
|
|
|
|
|
These are controlnet weights trained on stabilityai/stable-diffusion-xl-base-1.0 with OpenPose (v2) conditioning. You can find some example images in the following. |
|
|
|
prompt: a ballerina, romantic sunset, 4k photo |
|
![images_0)](./screenshot_ballerina.png) |
|
|
|
|
|
|
|
![images_0)](./out_ballerina.png) |
|
|
|
|
|
(Image is from ComfyUI, you can drag and drop in Comfy to use it as workflow) |
|
|
|
License: refers to the OpenPose's one. |
|
|
|
|
|
|
|
First, install all the libraries: |
|
|
|
```bash |
|
pip install -q controlnet_aux transformers accelerate |
|
pip install -q git+https://github.com/huggingface/diffusers |
|
``` |
|
|
|
Now, we're ready to make Darth Vader dance: |
|
|
|
```python |
|
from diffusers import AutoencoderKL, StableDiffusionXLControlNetPipeline, ControlNetModel, UniPCMultistepScheduler |
|
import torch |
|
from controlnet_aux import OpenposeDetector |
|
from diffusers.utils import load_image |
|
|
|
|
|
|
|
openpose = OpenposeDetector.from_pretrained("lllyasviel/ControlNet") |
|
|
|
image = load_image( |
|
"https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/person.png" |
|
) |
|
openpose_image = openpose(image) |
|
|
|
|
|
controlnet = ControlNetModel.from_pretrained("thibaud/controlnet-openpose-sdxl-1.0", torch_dtype=torch.float16) |
|
pipe = StableDiffusionXLControlNetPipeline.from_pretrained( |
|
"stabilityai/stable-diffusion-xl-base-1.0", controlnet=controlnet, torch_dtype=torch.float16 |
|
) |
|
pipe.enable_model_cpu_offload() |
|
|
|
|
|
|
|
prompt = "Darth vader dancing in a desert, high quality" |
|
negative_prompt = "low quality, bad quality" |
|
images = pipe( |
|
prompt, |
|
negative_prompt=negative_prompt, |
|
num_inference_steps=25, |
|
num_images_per_prompt=4, |
|
image=openpose_image.resize((1024, 1024)), |
|
generator=torch.manual_seed(97), |
|
).images |
|
images[0] |
|
``` |
|
|
|
Here are some gemerated examples: |
|
|
|
![](./darth_vader_grid.png) |
|
|
|
|
|
|
|
|
|
Use of the training script by HF🤗 [here](https://github.com/huggingface/diffusers/blob/main/examples/controlnet/README_sdxl.md). |
|
|
|
|
|
This checkpoint was first trained for 15,000 steps on laion 6a resized to a max minimum dimension of 768. |
|
|
|
|
|
one 1xA100 machine (Thanks a lot HF🤗 to provide the compute!) |
|
|
|
|
|
Data parallel with a single gpu batch size of 2 with gradient accumulation 8. |
|
|
|
|
|
Constant learning rate of 8e-5 |
|
|
|
|
|
fp16 |