Application: sd-sandbox: Stable Diffusion ONNX pipelines for AMD Ryzen™ AI Version: 1.9.0 (for AMD Ryzen™ AI Software 1.9.0) License: MIT for code written by Advanced Micro Devices, Inc. Third-party components are listed below with their own licenses. See LICENSE.
Every source file carries an SPDX license identifier or the full license header of its upstream project.
These files are derived from or taken from external projects. They keep the original copyright notice and license header. Where AMD changed a file, a Modifications Copyright Advanced Micro Devices, Inc. line was added and the upstream notice was left in place.
| File | Origin | Upstream copyright holder | License | AMD modified |
|---|---|---|---|---|
src/pipeline/pipeline_stable_diffusion_onnx_amd.py |
huggingface/diffusers | The HuggingFace Team | Apache-2.0 | Yes |
src/pipeline/pipeline_stable_diffusion_xl_onnx_amd.py |
huggingface/diffusers | The HuggingFace Team | Apache-2.0 | Yes |
src/pipeline/pipeline_controlnet_onnx_amd.py |
huggingface/diffusers | The HuggingFace Team | Apache-2.0 | Yes |
src/pipeline/pipeline_stable_diffusion_Kolors_onnx_amd.py |
huggingface/diffusers | The HuggingFace Team | Apache-2.0 | Yes |
src/pipeline/pipeline_stable_diffusion_3_controlnet_onnx_amd.py |
huggingface/diffusers | Stability AI, The HuggingFace Team and The InstantX Team | Apache-2.0 | Yes |
src/pipeline/pipeline_stable_diffusion_3_controlnet_outpainting_onnx_amd.py |
huggingface/diffusers | Stability AI, The HuggingFace Team and The AlimamaCreative Team | Apache-2.0 | Yes |
src/pipeline/pipeline_flux_onnx_amd.py |
huggingface/diffusers | Black Forest Labs and The HuggingFace Team | Apache-2.0 | Yes |
src/pipeline/pipeline_stable_diffusion_flux2_klein_onnx_amd.py |
huggingface/diffusers | Black Forest Labs and The HuggingFace Team | Apache-2.0 | Yes |
src/pipeline/pipeline_stable_diffusion_NitroE_onnx_amd.py |
huggingface/diffusers StableDiffusion3Pipeline (v0.31.0) |
Stability AI and The HuggingFace Team | Apache-2.0 | Yes |
src/flux2/image_processor.py |
huggingface/diffusers | The Black Forest Labs Team and The HuggingFace Team | Apache-2.0 | Yes |
src/t5/modeling_t5.py |
huggingface/transformers (models/t5/modeling_t5.py) |
Mesh TensorFlow authors, T5 Authors and HuggingFace Inc. team | Apache-2.0 | Yes |
src/kolors/configuration_chatglm.py |
Kwai-Kolors/Kolors @ 038818d | Kwai-Kolors (no holder named upstream) | Apache-2.0 | No |
src/kolors/modeling_chatglm.py |
Kwai-Kolors/Kolors @ 038818d | Kwai-Kolors (no holder named upstream) | Apache-2.0 | No |
src/kolors/tokenization_chatglm.py |
Kwai-Kolors/Kolors @ 038818d | Kwai-Kolors (no holder named upstream) | Apache-2.0 | Yes (_pad() padding_side argument) |
Note on the Nitro-E pipeline: the ONNX pipeline is adapted from the diffusers StableDiffusion3Pipeline (check_inputs, encode_prompt, and prepare_latents follow it), not from the AMD-AGI/Nitro-E training repository (MIT). Nitro-E's own diffusers pipeline (core/models/emmdit_pipeline.py) is also Apache-2.0, for the same reason.
All other source files (scripts/, src/runner/, src/trigger/, src/utils/, test/, and the remaining files under src/) were written by Advanced Micro Devices, Inc. and are MIT-licensed (SPDX-License-Identifier: MIT).
None. The repository contains no compiled binaries, libraries, or model weights. Models are downloaded at run time from Hugging Face or supplied by the user.
Dependencies are installed by the user through conda and pip. They are not redistributed in this repository. The authoritative, pinned lists are in the environment files:
| Environment file | Python | Platform |
|---|---|---|
env.yaml |
3.10 | Windows |
env_py312.yaml |
3.12 | Windows |
env_py312_vaip.yaml |
3.12 | Windows (Vitis AI EP) |
env_py313.yaml |
3.13 | Windows |
env_linux_py312.yaml |
3.12 | Linux |
Primary direct dependencies:
| Package | Version(s) pinned | License |
|---|---|---|
| diffusers | 0.31.0, 0.35.1 | Apache-2.0 |
| transformers | 4.42.4, 4.56.2, 4.57.0 | Apache-2.0 |
| torch | 2.4.1, 2.8.0 | BSD-3-Clause |
| onnxruntime / onnxruntime-directml | 1.21.0, 1.24.3 | MIT |
| onnxruntime-extensions | 0.12.0 | MIT |
| onnx | 1.16.1, 1.21.0 | Apache-2.0 |
| onnx-tool | 0.9.0 | MIT |
| optimum | 1.24.0, 2.0.0 | Apache-2.0 |
| accelerate | 0.32.0, 1.10.1 | Apache-2.0 |
| huggingface-hub | 0.23.4, 0.35.3, 0.36.0 | Apache-2.0 |
| safetensors | 0.4.3, 0.6.2 | Apache-2.0 |
| sentencepiece | 0.2.0, 0.2.1 | Apache-2.0 |
| numpy | 1.26.4, 2.3.3 | BSD-3-Clause |
| pillow | 10.4.0, 11.3.0 | MIT-CMU (HPND) |
| protobuf | 6.31.1, 6.33.5 | BSD-3-Clause |
| scipy | 1.15.1, 1.16.2 | BSD-3-Clause |
TODO: a complete license inventory of every transitive package in the environment files has not been produced yet.
Images used as test inputs, reference outputs, and README diagrams, with the origin and license of each. Files under test/ref/ that duplicate a file under test/assets/ are listed separately because they are separate files in the repository.
These images are redistributed unmodified from public upstream sources. The copyright holder and license of each source are retained here.
| File | Origin | Copyright holder | License | AMD modified |
|---|---|---|---|---|
test/assets/canny.jpg |
InstantX/SD3-Controlnet-Canny @ e42ecba canny.jpg |
The InstantX Team | Apache-2.0 (see note 1) | No |
test/ref/canny.jpg |
InstantX/SD3-Controlnet-Canny @ e42ecba canny.jpg |
The InstantX Team | Apache-2.0 (see note 1) | No |
test/assets/control.png |
lllyasviel/sd-controlnet-canny @ 7f2f691 images/bird_canny.png |
Lvmin Zhang (lllyasviel) | CreativeML OpenRAIL-M (see note 2) | No (same pixels, re-encoded) |
test/ref/control.png |
lllyasviel/sd-controlnet-canny @ 7f2f691 images/bird_canny.png |
Lvmin Zhang (lllyasviel) | CreativeML OpenRAIL-M (see note 2) | No (same pixels, re-encoded) |
test/assets/depth.jpeg |
InstantX/SD3-Controlnet-Depth @ 7765870 images/depth.jpeg |
The InstantX Team | Apache-2.0 | No |
test/ref/depth.jpeg |
InstantX/SD3-Controlnet-Depth @ 7765870 images/depth.jpeg |
The InstantX Team | Apache-2.0 | No |
test/assets/pose.jpg |
InstantX/SD3-Controlnet-Pose @ 1c550db pose.jpg |
The InstantX Team | Apache-2.0 (see note 1) | No |
test/ref/pose.jpg |
InstantX/SD3-Controlnet-Pose @ 1c550db pose.jpg |
The InstantX Team | Apache-2.0 (see note 1) | No |
test/assets/tile.jpg |
InstantX/SD3-Controlnet-Tile @ 48005f2 tile.jpg |
The InstantX Team | Apache-2.0 (see note 1) | No |
test/ref/tile.jpg |
InstantX/SD3-Controlnet-Tile @ 48005f2 tile.jpg |
The InstantX Team | Apache-2.0 (see note 1) | No |
test/ref/Canny_dog_condition.jpg |
Kwai-Kolors/Kolors @ 038818d controlnet/outputs/Canny_dog_condition.jpg |
Kuaishou Technology (Kwai-Kolors) | Apache-2.0 | No |
test/ref/inpainting/kolors_origin.png |
Kwai-Kolors/Kolors @ 038818d inpainting/asset/4.png |
Kuaishou Technology (Kwai-Kolors) | Apache-2.0 | No |
test/ref/inpainting/kolors_mask.png |
Kwai-Kolors/Kolors @ 038818d inpainting/asset/4_mask.png |
Kuaishou Technology (Kwai-Kolors) | Apache-2.0 | No |
| File | Origin | License |
|---|---|---|
test/assets/removal_input_1024x1024.jpg |
Provided to Advanced Micro Devices, Inc. by a partner for use in this project | Redistributed with the permission of the provider (see note 3) |
test/ref/removal/origin.jpg |
Copy of test/assets/removal_input_1024x1024.jpg |
Same as test/assets/removal_input_1024x1024.jpg |
test/assets/removal_input_mask_1024x1024.jpg |
Mask created by Advanced Micro Devices, Inc. for removal_input_1024x1024.jpg |
MIT |
test/ref/removal/mask.jpg |
Copy of test/assets/removal_input_mask_1024x1024.jpg |
MIT |
test/assets/removal_output_1024x1024.png |
Generated by Advanced Micro Devices, Inc. from removal_input_1024x1024.jpg and its mask |
MIT for the AMD contribution; includes content from the partner-provided input (see note 3) |
These images were generated or drawn by Advanced Micro Devices, Inc. and are distributed under the MIT license of this repository. Where a generative model was used, the model and its license are named. None of those licenses claim ownership of generated output.
| File | Origin | License |
|---|---|---|
test/assets/coffee_table.png |
Generated with SD 3.0 Medium by Advanced Micro Devices, Inc. (commit 3377d8a) | MIT |
test/assets/controlimg_input_1024x1024.png |
Generated by Advanced Micro Devices, Inc. with Segmind Vega (Apache-2.0) | MIT |
test/assets/controlimg_output_strength_0.95_1024x1024.png |
Generated by Advanced Micro Devices, Inc. with Segmind Vega (Apache-2.0) | MIT |
test/assets/flux2_img1.png |
Generated by Advanced Micro Devices, Inc. with FLUX.2 klein 4B (Apache-2.0) | MIT |
test/assets/flux2_img2.png |
Generated by Advanced Micro Devices, Inc. with FLUX.2 klein 4B (Apache-2.0) | MIT |
test/assets/flux2_img3.png |
Generated by Advanced Micro Devices, Inc. with FLUX.2 klein 4B (Apache-2.0) | MIT |
test/assets/image-to-image-controlnet-workflow.png |
Original diagram by Advanced Micro Devices, Inc. (commit 3377d8a) | MIT |
test/assets/image-to-image-workflow.png |
Original diagram by Advanced Micro Devices, Inc. (commit bead652) | MIT |
test/assets/inpainting_output_1024x1024.png |
Generated with SD 3.0 Medium by Advanced Micro Devices, Inc. (commit 3377d8a) | MIT |
test/assets/living_room_mask_books_vase_1024x1024.png |
Generated or created by Advanced Micro Devices, Inc. with AI models running on AMD hardware | MIT |
test/assets/outpainting_input_512x512.png |
Generated by Advanced Micro Devices, Inc. with Stable Diffusion 3 Medium (Stability AI Community License) | MIT |
test/assets/outpainting_output_1024x1024.png |
Generated or created by Advanced Micro Devices, Inc. with AI models running on AMD hardware | MIT |
test/assets/precise_mask.png |
Created by Advanced Micro Devices, Inc. to match coffee_table.png (commit 3377d8a) |
MIT |
test/assets/sd_demo.gif |
Generated or created by Advanced Micro Devices, Inc. with AI models running on AMD hardware | MIT |
test/assets/text-to-image-workflow.png |
Original diagram by Advanced Micro Devices, Inc. (commit 3377d8a) | MIT |
test/ref/inpainting/mask.jpg |
Generated or created by Advanced Micro Devices, Inc. with AI models running on AMD hardware | MIT |
test/ref/inpainting/origin.jpg |
Generated or created by Advanced Micro Devices, Inc. with AI models running on AMD hardware | MIT |
test/ref/outpainting.jpg |
Generated or created by Advanced Micro Devices, Inc. with AI models running on AMD hardware | MIT |
test/ref/union.jfif |
Generated or created by Advanced Micro Devices, Inc. with AI models running on AMD hardware | MIT |
Notes:
- The InstantX SD3 ControlNet repositories are published as a set, and
SD3-Controlnet-Depthis licensed Apache-2.0. TheSD3-Controlnet-Canny,SD3-Controlnet-Pose, andSD3-Controlnet-Tilemodel cards do not repeat the license, so their sample images are treated as Apache-2.0 on the same basis. images/bird_canny.pngis published in thelllyasviel/sd-controlnet-cannyrepository, which is licensed under CreativeML OpenRAIL-M. The use-based restrictions of that license continue to apply to this image.- This image was supplied to Advanced Micro Devices, Inc. by a partner for use in this project. It is not covered by the MIT license of this repository.
None. Every file in this repository is intended for public release.