PiSA-SR ONNX models for AnyImage

PiSA-SR: Lingchen Sun and contributors, https://github.com/csslc/PiSA-SR
Source commit: 9c76e12e261d1980148738a0cff41e5e27d18c9f
PiSA-SR is released under Apache License 2.0; see LICENSE-APACHE.txt.

Stable Diffusion 2.1: Copyright (c) 2022 Stability AI and contributors.
Base weights: sd2-community/stable-diffusion-2-1-base
Revision: 4e63672c03103b6c636b8fb4119ba982469b2955
Base weights and derivatives are subject to CreativeML Open RAIL++-M,
including paragraph 5 and Attachment A use restrictions. All users must
comply with these terms; see the accompanying LICENSE-MODEL.txt.
License copy: sd2-community/stable-diffusion-2, LICENSE-MODEL.

AnyImage modifications: merged pixel and semantic LoRA adapters, FP16
ONNX conversion, fixed empty prompt and timestep, VAE encoder/decoder
separation. Runtime combines pixel/full predictions at semantic strength 0.7.
noise.npy contains a fixed seed 42 latent noise field.

ONNX architecture derives from Hugging Face Diffusers and PiSA-SR.
Copyright 2023 The HuggingFace Team. All rights reserved.
