Stable-Diffusion-v2.1: Optimized for Qualcomm Devices

Generates high resolution images from text prompts using a latent diffusion model. This model uses CLIP ViT-L/14 as text encoder, U-Net based latent denoising, and VAE based decoder to generate the final image.

This is based on the implementation of Stable-Diffusion-v2.1 found here. This repository contains pre-exported model files optimized for Qualcomm® devices. You can use the Qualcomm® AI Hub Models library to export with custom configurations. More details on model performance across various devices, can be found here.

Qualcomm AI Hub Models uses Qualcomm AI Hub Workbench to compile, profile, and evaluate this model. Sign up to run these models on a hosted Qualcomm® device.

Deploy to Snapdragon X Elite NPU

Please follow the Stable Diffusion Windows App tutorial to quantize model with custom weights.

Quantize and Deploy Your Own Fine-Tuned Stable Diffusion

Please follow the Quantize Stable Diffusion tutorial to quantize model with custom weights.

Getting Started

There are two ways to deploy this model on your device:

Option 1: Download Pre-Exported Models

Below are pre-exported model assets ready for deployment.

Runtime	Precision	Chipset	SDK Versions	Download
PRECOMPILED_QNN_ONNX	w8a16	Snapdragon® X Elite	QAIRT 2.37, ONNX Runtime 1.23.0	Download
PRECOMPILED_QNN_ONNX	w8a16	Snapdragon® 8 Gen 3 Mobile	QAIRT 2.37, ONNX Runtime 1.23.0	Download
PRECOMPILED_QNN_ONNX	w8a16	Qualcomm® QCS8550 (Proxy)	QAIRT 2.37, ONNX Runtime 1.23.0	Download
PRECOMPILED_QNN_ONNX	w8a16	Snapdragon® 8 Elite For Galaxy Mobile	QAIRT 2.37, ONNX Runtime 1.23.0	Download
PRECOMPILED_QNN_ONNX	w8a16	Snapdragon® 7 Gen 4 Mobile	QAIRT 2.37, ONNX Runtime 1.23.0	Download
PRECOMPILED_QNN_ONNX	w8a16	Snapdragon® 8 Elite Gen 5 Mobile	QAIRT 2.37, ONNX Runtime 1.23.0	Download
PRECOMPILED_QNN_ONNX	w8a16	Qualcomm® QCM6690	QAIRT 2.37, ONNX Runtime 1.23.0	Download
PRECOMPILED_QNN_ONNX	w8a16	Qualcomm® QCS9075	QAIRT 2.37, ONNX Runtime 1.23.0	Download
QNN_CONTEXT_BINARY	w8a16	Snapdragon® X Elite	QAIRT 2.42	Download
QNN_CONTEXT_BINARY	w8a16	Snapdragon® 8 Gen 3 Mobile	QAIRT 2.42	Download
QNN_CONTEXT_BINARY	w8a16	Qualcomm® QCS8550 (Proxy)	QAIRT 2.42	Download
QNN_CONTEXT_BINARY	w8a16	Qualcomm® SA8775P	QAIRT 2.42	Download
QNN_CONTEXT_BINARY	w8a16	Snapdragon® 8 Elite For Galaxy Mobile	QAIRT 2.42	Download
QNN_CONTEXT_BINARY	w8a16	Snapdragon® 7 Gen 4 Mobile	QAIRT 2.42	Download
QNN_CONTEXT_BINARY	w8a16	Snapdragon® 8 Elite Gen 5 Mobile	QAIRT 2.42	Download
QNN_CONTEXT_BINARY	w8a16	Qualcomm® QCM6690	QAIRT 2.42	Download
QNN_CONTEXT_BINARY	w8a16	Qualcomm® QCS9075	QAIRT 2.42	Download

For more device-specific assets and performance metrics, visit Stable-Diffusion-v2.1 on Qualcomm® AI Hub.

Option 2: Export with Custom Configurations

Use the Qualcomm® AI Hub Models Python library to compile and export the model with your own:

• Custom weights (e.g., fine-tuned checkpoints)
• Custom input shapes
• Target device and runtime configurations

This option is ideal if you need to customize the model beyond the default configuration provided here.

See our repository for Stable-Diffusion-v2.1 on GitHub for usage instructions.

Model Details

Model Type: Model_use_case.image_generation

Model Stats:

• Input: Text prompt to generate image

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
text_encoder	PRECOMPILED_QNN_ONNX	w8a16	Snapdragon® X Elite	7.792 ms	378 - 378 MB	NPU
text_encoder	PRECOMPILED_QNN_ONNX	w8a16	Snapdragon® 8 Gen 3 Mobile	5.109 ms	0 - 12 MB	NPU
text_encoder	PRECOMPILED_QNN_ONNX	w8a16	Qualcomm® QCS8550 (Proxy)	7.751 ms	0 - 386 MB	NPU
text_encoder	PRECOMPILED_QNN_ONNX	w8a16	Qualcomm® QCS9075	7.588 ms	0 - 3 MB	NPU
text_encoder	PRECOMPILED_QNN_ONNX	w8a16	Qualcomm® QCM6690	67.98 ms	0 - 11 MB	NPU
text_encoder	PRECOMPILED_QNN_ONNX	w8a16	Snapdragon® 8 Elite For Galaxy Mobile	4.435 ms	0 - 12 MB	NPU
text_encoder	PRECOMPILED_QNN_ONNX	w8a16	Snapdragon® 7 Gen 4 Mobile	11.287 ms	0 - 7 MB	NPU
text_encoder	PRECOMPILED_QNN_ONNX	w8a16	Snapdragon® 8 Elite Gen 5 Mobile	4.234 ms	0 - 10 MB	NPU
text_encoder	QNN_CONTEXT_BINARY	w8a16	Snapdragon® X Elite	8.348 ms	0 - 0 MB	NPU
text_encoder	QNN_CONTEXT_BINARY	w8a16	Snapdragon® 8 Gen 3 Mobile	5.602 ms	0 - 7 MB	NPU
text_encoder	QNN_CONTEXT_BINARY	w8a16	Qualcomm® QCS8550 (Proxy)	7.958 ms	0 - 2 MB	NPU
text_encoder	QNN_CONTEXT_BINARY	w8a16	Qualcomm® SA8775P	8.369 ms	0 - 9 MB	NPU
text_encoder	QNN_CONTEXT_BINARY	w8a16	Qualcomm® QCS9075	7.967 ms	0 - 2 MB	NPU
text_encoder	QNN_CONTEXT_BINARY	w8a16	Qualcomm® QCM6690	76.001 ms	0 - 6 MB	NPU
text_encoder	QNN_CONTEXT_BINARY	w8a16	Snapdragon® 8 Elite For Galaxy Mobile	4.632 ms	0 - 12 MB	NPU
text_encoder	QNN_CONTEXT_BINARY	w8a16	Snapdragon® 7 Gen 4 Mobile	11.506 ms	0 - 6 MB	NPU
text_encoder	QNN_CONTEXT_BINARY	w8a16	Snapdragon® 8 Elite Gen 5 Mobile	4.288 ms	0 - 9 MB	NPU
unet	PRECOMPILED_QNN_ONNX	w8a16	Snapdragon® X Elite	95.816 ms	843 - 843 MB	NPU
unet	PRECOMPILED_QNN_ONNX	w8a16	Snapdragon® 8 Gen 3 Mobile	69.512 ms	0 - 13 MB	NPU
unet	PRECOMPILED_QNN_ONNX	w8a16	Qualcomm® QCS8550 (Proxy)	95.789 ms	0 - 898 MB	NPU
unet	PRECOMPILED_QNN_ONNX	w8a16	Qualcomm® QCS9075	88.938 ms	0 - 3 MB	NPU
unet	PRECOMPILED_QNN_ONNX	w8a16	Qualcomm® QCM6690	2057.646 ms	0 - 11 MB	NPU
unet	PRECOMPILED_QNN_ONNX	w8a16	Snapdragon® 8 Elite For Galaxy Mobile	56.546 ms	1 - 13 MB	NPU
unet	PRECOMPILED_QNN_ONNX	w8a16	Snapdragon® 7 Gen 4 Mobile	152.469 ms	0 - 10 MB	NPU
unet	PRECOMPILED_QNN_ONNX	w8a16	Snapdragon® 8 Elite Gen 5 Mobile	39.323 ms	0 - 8 MB	NPU
unet	QNN_CONTEXT_BINARY	w8a16	Snapdragon® X Elite	96.293 ms	0 - 0 MB	NPU
unet	QNN_CONTEXT_BINARY	w8a16	Snapdragon® 8 Gen 3 Mobile	69.688 ms	0 - 7 MB	NPU
unet	QNN_CONTEXT_BINARY	w8a16	Qualcomm® QCS8550 (Proxy)	95.825 ms	0 - 2 MB	NPU
unet	QNN_CONTEXT_BINARY	w8a16	Qualcomm® SA8775P	90.588 ms	0 - 7 MB	NPU
unet	QNN_CONTEXT_BINARY	w8a16	Qualcomm® QCS9075	89.813 ms	0 - 2 MB	NPU
unet	QNN_CONTEXT_BINARY	w8a16	Qualcomm® QCM6690	2236.027 ms	0 - 6 MB	NPU
unet	QNN_CONTEXT_BINARY	w8a16	Snapdragon® 8 Elite For Galaxy Mobile	53.213 ms	0 - 13 MB	NPU
unet	QNN_CONTEXT_BINARY	w8a16	Snapdragon® 7 Gen 4 Mobile	178.363 ms	0 - 6 MB	NPU
unet	QNN_CONTEXT_BINARY	w8a16	Snapdragon® 8 Elite Gen 5 Mobile	39.59 ms	0 - 9 MB	NPU
vae	PRECOMPILED_QNN_ONNX	w8a16	Snapdragon® X Elite	217.654 ms	58 - 58 MB	NPU
vae	PRECOMPILED_QNN_ONNX	w8a16	Snapdragon® 8 Gen 3 Mobile	162.419 ms	3 - 11 MB	NPU
vae	PRECOMPILED_QNN_ONNX	w8a16	Qualcomm® QCS8550 (Proxy)	218.196 ms	5 - 70 MB	NPU
vae	PRECOMPILED_QNN_ONNX	w8a16	Qualcomm® QCS9075	206.367 ms	4 - 6 MB	NPU
vae	PRECOMPILED_QNN_ONNX	w8a16	Qualcomm® QCM6690	5361.104 ms	0 - 7 MB	NPU
vae	PRECOMPILED_QNN_ONNX	w8a16	Snapdragon® 8 Elite For Galaxy Mobile	145.886 ms	4 - 15 MB	NPU
vae	PRECOMPILED_QNN_ONNX	w8a16	Snapdragon® 7 Gen 4 Mobile	446.615 ms	3 - 10 MB	NPU
vae	PRECOMPILED_QNN_ONNX	w8a16	Snapdragon® 8 Elite Gen 5 Mobile	89.465 ms	3 - 14 MB	NPU
vae	QNN_CONTEXT_BINARY	w8a16	Snapdragon® X Elite	264.139 ms	0 - 0 MB	NPU
vae	QNN_CONTEXT_BINARY	w8a16	Snapdragon® 8 Gen 3 Mobile	201.927 ms	0 - 7 MB	NPU
vae	QNN_CONTEXT_BINARY	w8a16	Qualcomm® QCS8550 (Proxy)	266.775 ms	0 - 2 MB	NPU
vae	QNN_CONTEXT_BINARY	w8a16	Qualcomm® SA8775P	1136.127 ms	0 - 9 MB	NPU
vae	QNN_CONTEXT_BINARY	w8a16	Qualcomm® QCS9075	249.481 ms	0 - 3 MB	NPU
vae	QNN_CONTEXT_BINARY	w8a16	Qualcomm® QCM6690	6292.624 ms	0 - 6 MB	NPU
vae	QNN_CONTEXT_BINARY	w8a16	Snapdragon® 8 Elite For Galaxy Mobile	178.324 ms	0 - 9 MB	NPU
vae	QNN_CONTEXT_BINARY	w8a16	Snapdragon® 7 Gen 4 Mobile	511.921 ms	1 - 7 MB	NPU
vae	QNN_CONTEXT_BINARY	w8a16	Snapdragon® 8 Elite Gen 5 Mobile	118.903 ms	0 - 10 MB	NPU

License

• The license for the original implementation of Stable-Diffusion-v2.1 can be found here.

References

• High-Resolution Image Synthesis with Latent Diffusion Models
• Source Model Implementation

Community

• Join our AI Hub Slack community to collaborate, post questions and learn more about on-device AI.
• For questions or feedback please reach out to us.

Usage and Limitations

This model may not be used for or in connection with any of the following applications:

• Accessing essential private and public services and benefits;
• Administration of justice and democratic processes;
• Assessing or recognizing the emotional state of a person;
• Biometric and biometrics-based systems, including categorization of persons based on sensitive characteristics;
• Education and vocational training;
• Employment and workers management;
• Exploitation of the vulnerabilities of persons resulting in harmful behavior;
• General purpose social scoring;
• Law enforcement;
• Management and operation of critical infrastructure;
• Migration, asylum and border control management;
• Predictive policing;
• Real-time remote biometric identification in public spaces;
• Recommender systems of social media platforms;
• Scraping of facial images (from the internet or otherwise); and/or
• Subliminal manipulation