|
--- |
|
license: openrail++ |
|
language: |
|
- en |
|
tags: |
|
- stablediffusion |
|
- openvino |
|
--- |
|
HyperSD 1-step LoRA baked-in model converted to OpenVINO(int8) and weights compressed with NNCF(10GB to 4.4GB). |
|
|
|
Original Model : [Hyper-SD](https://huggingface.co/ByteDance/Hyper-SD) |
|
|
|
You can use this model with [FastSD CPU](https://github.com/rupeshs/fastsdcpu). |
|
|
|
 |
|
|
|
To run the model yourself, you can leverage the 🧨 Diffusers library: |
|
|
|
1. Install the dependencies: |
|
``` |
|
pip install optimum-intel openvino diffusers onnx |
|
``` |
|
2. Run the model: |
|
```py |
|
from optimum.intel.openvino.modeling_diffusion import OVStableDiffusionXLPipeline |
|
|
|
pipeline = OVStableDiffusionXLPipeline.from_pretrained( |
|
"rupeshs/hyper-sd-sdxl-1-step-openvino-int8", |
|
ov_config={"CACHE_DIR": ""}, |
|
) |
|
prompt = "a cute cat,flowers" |
|
|
|
images = pipeline( |
|
prompt=prompt, |
|
width=768, |
|
height=768, |
|
num_inference_steps=1, |
|
guidance_scale=1.0, |
|
).images |
|
images[0].save("out_image.png") |
|
``` |
|
|
