--- library_name: pytorch license: other tags: - generative_ai - android pipeline_tag: unconditional-image-generation --- ![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/stable_diffusion_v2_1/web-assets/model_demo.png) # Stable-Diffusion-v2.1: Optimized for Mobile Deployment ## State-of-the-art generative AI model used to generate detailed images conditioned on text descriptions 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 model is an implementation of Stable-Diffusion-v2.1 found [here](https://github.com/CompVis/stable-diffusion/tree/main). This repository provides scripts to run Stable-Diffusion-v2.1 on Qualcomm® devices. More details on model performance across various devices, can be found [here](https://aihub.qualcomm.com/models/stable_diffusion_v2_1). ### Model Details - **Model Type:** Model_use_case.image_generation - **Model Stats:** - Input: Text prompt to generate image | Model | Precision | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit | Target Model |---|---|---|---|---|---|---|---|---| | text_encoder | w8a16 | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN_CONTEXT_BINARY | 18.499 ms | 0 - 8 MB | NPU | Use Export Script | | text_encoder | w8a16 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN_CONTEXT_BINARY | 8.205 ms | 0 - 3 MB | NPU | Use Export Script | | text_encoder | w8a16 | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN_CONTEXT_BINARY | 8.377 ms | 0 - 9 MB | NPU | Use Export Script | | text_encoder | w8a16 | SA7255P ADP | Qualcomm® SA7255P | QNN_CONTEXT_BINARY | 18.499 ms | 0 - 8 MB | NPU | Use Export Script | | text_encoder | w8a16 | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | QNN_CONTEXT_BINARY | 8.366 ms | 0 - 2 MB | NPU | Use Export Script | | text_encoder | w8a16 | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | QNN_CONTEXT_BINARY | 8.216 ms | 0 - 2 MB | NPU | Use Export Script | | text_encoder | w8a16 | SA8775P ADP | Qualcomm® SA8775P | QNN_CONTEXT_BINARY | 8.377 ms | 0 - 9 MB | NPU | Use Export Script | | text_encoder | w8a16 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | QNN_CONTEXT_BINARY | 8.285 ms | 0 - 7 MB | NPU | Use Export Script | | text_encoder | w8a16 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | PRECOMPILED_QNN_ONNX | 8.141 ms | 0 - 387 MB | NPU | Use Export Script | | text_encoder | w8a16 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN_CONTEXT_BINARY | 5.601 ms | 0 - 18 MB | NPU | Use Export Script | | text_encoder | w8a16 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | PRECOMPILED_QNN_ONNX | 5.404 ms | 0 - 19 MB | NPU | Use Export Script | | text_encoder | w8a16 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | QNN_CONTEXT_BINARY | 4.56 ms | 0 - 15 MB | NPU | Use Export Script | | text_encoder | w8a16 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | PRECOMPILED_QNN_ONNX | 5.312 ms | 0 - 14 MB | NPU | Use Export Script | | text_encoder | w8a16 | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN_CONTEXT_BINARY | 8.652 ms | 0 - 0 MB | NPU | Use Export Script | | text_encoder | w8a16 | Snapdragon X Elite CRD | Snapdragon® X Elite | PRECOMPILED_QNN_ONNX | 8.311 ms | 378 - 378 MB | NPU | Use Export Script | | unet | w8a16 | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN_CONTEXT_BINARY | 232.583 ms | 0 - 8 MB | NPU | Use Export Script | | unet | w8a16 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN_CONTEXT_BINARY | 93.286 ms | 0 - 2 MB | NPU | Use Export Script | | unet | w8a16 | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN_CONTEXT_BINARY | 87.78 ms | 0 - 9 MB | NPU | Use Export Script | | unet | w8a16 | SA7255P ADP | Qualcomm® SA7255P | QNN_CONTEXT_BINARY | 232.583 ms | 0 - 8 MB | NPU | Use Export Script | | unet | w8a16 | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | QNN_CONTEXT_BINARY | 93.785 ms | 0 - 3 MB | NPU | Use Export Script | | unet | w8a16 | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | QNN_CONTEXT_BINARY | 93.873 ms | 0 - 3 MB | NPU | Use Export Script | | unet | w8a16 | SA8775P ADP | Qualcomm® SA8775P | QNN_CONTEXT_BINARY | 87.78 ms | 0 - 9 MB | NPU | Use Export Script | | unet | w8a16 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | QNN_CONTEXT_BINARY | 93.717 ms | 0 - 7 MB | NPU | Use Export Script | | unet | w8a16 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | PRECOMPILED_QNN_ONNX | 95.033 ms | 0 - 898 MB | NPU | Use Export Script | | unet | w8a16 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN_CONTEXT_BINARY | 66.248 ms | 0 - 18 MB | NPU | Use Export Script | | unet | w8a16 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | PRECOMPILED_QNN_ONNX | 67.516 ms | 0 - 14 MB | NPU | Use Export Script | | unet | w8a16 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | QNN_CONTEXT_BINARY | 58.258 ms | 0 - 14 MB | NPU | Use Export Script | | unet | w8a16 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | PRECOMPILED_QNN_ONNX | 58.617 ms | 0 - 19 MB | NPU | Use Export Script | | unet | w8a16 | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN_CONTEXT_BINARY | 95.326 ms | 0 - 0 MB | NPU | Use Export Script | | unet | w8a16 | Snapdragon X Elite CRD | Snapdragon® X Elite | PRECOMPILED_QNN_ONNX | 96.279 ms | 843 - 843 MB | NPU | Use Export Script | | vae | w8a16 | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN_CONTEXT_BINARY | 719.113 ms | 0 - 9 MB | NPU | Use Export Script | | vae | w8a16 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN_CONTEXT_BINARY | 270.059 ms | 1 - 3 MB | NPU | Use Export Script | | vae | w8a16 | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN_CONTEXT_BINARY | 249.046 ms | 0 - 10 MB | NPU | Use Export Script | | vae | w8a16 | SA7255P ADP | Qualcomm® SA7255P | QNN_CONTEXT_BINARY | 719.113 ms | 0 - 9 MB | NPU | Use Export Script | | vae | w8a16 | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | QNN_CONTEXT_BINARY | 271.52 ms | 0 - 3 MB | NPU | Use Export Script | | vae | w8a16 | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | QNN_CONTEXT_BINARY | 271.625 ms | 0 - 4 MB | NPU | Use Export Script | | vae | w8a16 | SA8775P ADP | Qualcomm® SA8775P | QNN_CONTEXT_BINARY | 249.046 ms | 0 - 10 MB | NPU | Use Export Script | | vae | w8a16 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | QNN_CONTEXT_BINARY | 269.852 ms | 0 - 2 MB | NPU | Use Export Script | | vae | w8a16 | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | PRECOMPILED_QNN_ONNX | 274.742 ms | 0 - 68 MB | NPU | Use Export Script | | vae | w8a16 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN_CONTEXT_BINARY | 202.714 ms | 0 - 21 MB | NPU | Use Export Script | | vae | w8a16 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | PRECOMPILED_QNN_ONNX | 203.279 ms | 3 - 22 MB | NPU | Use Export Script | | vae | w8a16 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | QNN_CONTEXT_BINARY | 175.114 ms | 0 - 14 MB | NPU | Use Export Script | | vae | w8a16 | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | PRECOMPILED_QNN_ONNX | 174.448 ms | 3 - 17 MB | NPU | Use Export Script | | vae | w8a16 | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN_CONTEXT_BINARY | 265.137 ms | 0 - 0 MB | NPU | Use Export Script | | vae | w8a16 | Snapdragon X Elite CRD | Snapdragon® X Elite | PRECOMPILED_QNN_ONNX | 264.748 ms | 62 - 62 MB | NPU | Use Export Script | ## Deploy to Snapdragon X Elite NPU Please follow the [Stable Diffusion Windows App](https://github.com/quic/ai-hub-apps/tree/main/apps/windows/python/StableDiffusion) tutorial to quantize model with custom weights. ## Quantize and Deploy Your Own Fine-Tuned Stable Diffusion Please follow the [Quantize Stable Diffusion]({REPOSITORY_URL}/tutorials/stable_diffusion/quantize_stable_diffusion.md) tutorial to quantize model with custom weights. ## Installation Install the package via pip: ```bash pip install "qai-hub-models[stable-diffusion-v2-1]" ``` ## Configure Qualcomm® AI Hub to run this model on a cloud-hosted device Sign-in to [Qualcomm® AI Hub](https://app.aihub.qualcomm.com/) with your Qualcomm® ID. Once signed in navigate to `Account -> Settings -> API Token`. With this API token, you can configure your client to run models on the cloud hosted devices. ```bash qai-hub configure --api_token API_TOKEN ``` Navigate to [docs](https://app.aihub.qualcomm.com/docs/) for more information. ## Demo off target The package contains a simple end-to-end demo that downloads pre-trained weights and runs this model on a sample input. ```bash python -m qai_hub_models.models.stable_diffusion_v2_1.demo ``` The above demo runs a reference implementation of pre-processing, model inference, and post processing. **NOTE**: If you want running in a Jupyter Notebook or Google Colab like environment, please add the following to your cell (instead of the above). ``` %run -m qai_hub_models.models.stable_diffusion_v2_1.demo ``` ### Run model on a cloud-hosted device In addition to the demo, you can also run the model on a cloud-hosted Qualcomm® device. This script does the following: * Performance check on-device on a cloud-hosted device * Downloads compiled assets that can be deployed on-device for Android. * Accuracy check between PyTorch and on-device outputs. ```bash python -m qai_hub_models.models.stable_diffusion_v2_1.export ``` ``` Profiling Results ------------------------------------------------------------ text_encoder Device : cs_8275 (ANDROID 14) Runtime : QNN_CONTEXT_BINARY Estimated inference time (ms) : 18.5 Estimated peak memory usage (MB): [0, 8] Total # Ops : 788 Compute Unit(s) : npu (788 ops) gpu (0 ops) cpu (0 ops) ------------------------------------------------------------ unet Device : cs_8275 (ANDROID 14) Runtime : QNN_CONTEXT_BINARY Estimated inference time (ms) : 232.6 Estimated peak memory usage (MB): [0, 8] Total # Ops : 5784 Compute Unit(s) : npu (5784 ops) gpu (0 ops) cpu (0 ops) ------------------------------------------------------------ vae Device : cs_8275 (ANDROID 14) Runtime : QNN_CONTEXT_BINARY Estimated inference time (ms) : 719.1 Estimated peak memory usage (MB): [0, 9] Total # Ops : 175 Compute Unit(s) : npu (175 ops) gpu (0 ops) cpu (0 ops) ``` ## Deploying compiled model to Android The models can be deployed using multiple runtimes: - TensorFlow Lite (`.tflite` export): [This tutorial](https://www.tensorflow.org/lite/android/quickstart) provides a guide to deploy the .tflite model in an Android application. - QNN (`.so` export ): This [sample app](https://docs.qualcomm.com/bundle/publicresource/topics/80-63442-50/sample_app.html) provides instructions on how to use the `.so` shared library in an Android application. ## View on Qualcomm® AI Hub Get more details on Stable-Diffusion-v2.1's performance across various devices [here](https://aihub.qualcomm.com/models/stable_diffusion_v2_1). Explore all available models on [Qualcomm® AI Hub](https://aihub.qualcomm.com/) ## License * The license for the original implementation of Stable-Diffusion-v2.1 can be found [here](https://github.com/CompVis/stable-diffusion/blob/main/LICENSE). * The license for the compiled assets for on-device deployment can be found [here](https://github.com/CompVis/stable-diffusion/blob/main/LICENSE) ## References * [High-Resolution Image Synthesis with Latent Diffusion Models](https://arxiv.org/abs/2112.10752) * [Source Model Implementation](https://github.com/CompVis/stable-diffusion/tree/main) ## Community * Join [our AI Hub Slack community](https://aihub.qualcomm.com/community/slack) to collaborate, post questions and learn more about on-device AI. * For questions or feedback please [reach out to us](mailto:ai-hub-support@qti.qualcomm.com).