tlwu/stable-diffusion-v1-5

Stable diffusion 1.5 ONNX models optimized for ONNX Runtime CUDA Execution Provider.

Installation

pip install onnxruntime-gpu>=1.14 diffusers>=0.13.0 transformers accelerate

onnxruntime-gpu requires CUDA and cuDNN. See https://onnxruntime.ai/docs/execution-providers/CUDA-ExecutionProvider.html for more information.

Stable Diffusion Inference

The snippet below demonstrates how to use the ONNX runtime to inference Stable diffusion 1.5 in Nvidia GPU:

# make sure you're logged in with `huggingface-cli login`
from diffusers import OnnxStableDiffusionPipeline

pipe = OnnxStableDiffusionPipeline.from_pretrained(
    "tlwu/stable-diffusion-v1-5",
    revision="fp16",
    provider="CUDAExecutionProvider",
)

prompt = "a photo of an astronaut riding a horse on mars"
image = pipe(prompt).images[0]
image.save("astronaut_rides_horse.png")

See this doc for more information about how the models are generated and the benchmark results.