It should work for GitHub - bes-dev/stable_diffusion.openvino at sb/onnxruntime, but I didn't test it.
Note: Not compatible with diffusers's ONNX pipeline.
In a private project I wrote myself, unet.onnx and vae_decoder.onnx are tested and confirmed to work.
Converted from the pre-trained weights wd-v1-3-full.ckpt in this repository: hakurei/waifu-diffusion-v1-3 路 Hugging Face.
The text encoder is from the original repository: bes-dev/stable-diffusion-v1-4-onnx 路 Hugging Face.
The computation graph for onnx in this repository is the same as in this repository: