intel · gblong1 · Sep 20, 2024 · Feb 9, 2023 · Feb 9, 2023 · Feb 10, 2023
diff --git a/Docs/figs/gimp_load_model.png b/Docs/figs/gimp_load_model.png
diff --git a/Docs/figs/gimp_sd_ui.png b/Docs/figs/gimp_sd_ui.png
diff --git a/Docs/figs/stable-diffusion_model1.png b/Docs/figs/stable-diffusion_model1.png
diff --git a/Docs/figs/stable-diffusion_model2.png b/Docs/figs/stable-diffusion_model2.png
diff --git a/Docs/figs/stable-diffusion_model3.png b/Docs/figs/stable-diffusion_model3.png
diff --git a/Docs/figs/stable-diffusion_model4.png b/Docs/figs/stable-diffusion_model4.png
diff --git a/Docs/figs/standalone1.png b/Docs/figs/standalone1.png
diff --git a/Docs/figs/standalone2.png b/Docs/figs/standalone2.png
diff --git a/Docs/stable-diffusion-v3.md b/Docs/stable-diffusion-v3.md
@@ -0,0 +1,38 @@
+# Image generation with Stable Diffusion 3.0
+
+Stable Diffusion V3 is next generation of latent diffusion image Stable Diffusion models family that outperforms state-of-the-art text-to-image generation systems in typography and prompt adherence, based on human preference evaluations. 
+
+In comparison with previous versions, it based on Multimodal Diffusion Transformer (MMDiT) text-to-image model that features greatly improved performance in image quality, typography, complex prompt understanding, and resource-efficiency.
+
+More details about model can be found in [model card](https://huggingface.co/stabilityai/stable-diffusion-3-medium), [research paper](https://stability.ai/news/stable-diffusion-3-research-paper) and [Stability.AI blog post](https://stability.ai/news/stable-diffusion-3-medium).
+
+## Enabling SD3 in GIMP
+### Converting the Models
+The supported version of SD3 in GIMP is the flash_sd3 LoRA. In order to use with GIMP, you first need to download and convert the model into OpenVINO format. This can be done by following the instructions in the [OpenVINO Stable Diffusion v3 Notebook](https://github.com/openvinotoolkit/openvino_notebooks/tree/latest/notebooks/stable-diffusion-v3).
+>**Note**: Ensure you use the instructions for flash_sd3. 
+
+### Model Installation
+After completing the steps in the notebook, you will have the models located in `openvino_notebooks\notebooks\stable-diffusion-v3\stable-diffusion-3`. Copy these models to `openvino-ai-plugins-gimp\weights\stable-diffusion-ov\stable-diffusion-3.0` 
+
+For example, using robocopy to copy the files on the command line:
+```
+robocopy openvino_notebooks\notebooks\stable-diffusion-v3\stable-diffusion-3\ %userprofile%\openvino-ai-plugins-gimp\weights\stable-diffusion-ov\stable-diffusion-3.0\. /mir
+```
+After copying, create a file called  `install_info.json` inside the stable-diffusion-3.0 directory. This will ensure that the GIMP plugin will recognize this as a valid model. It should contain the following text:
+```
+{
+    "hf_repo_id": "none",
+    "hf_commit_id": "none"
+}
+```
+
+Verify that the copied files and directory structure looks as follows:
+
+![image](https://github.com/user-attachments/assets/039073d1-e593-4365-92c9-3555ea023670)
+### Running with GIMP
+After completing model installation steps, SD3 will now be available in the Stable Diffusion UI. Note that SD3 can generate valid images in as few as 4 iterations. Also, the guidance scale needed is normally much lower. See the screenshot below:
+![image](https://github.com/user-attachments/assets/6daf3201-b873-4198-a752-19594f352c50)
+
+
+
+
diff --git a/Docs/user_guide_for_windows_users.md b/Docs/user_guide_for_windows_users.md
@@ -105,8 +105,7 @@ Please download [gimp-2.99.16-setup.exe](https://download.gimp.org/gimp/v2.99/wi
     cd C:\Users\Public\
     mkdir GIMP
     cd GIMP
-
-    git clone https://github.com/intel/openvino-ai-plugins-gimp.git
+    git clone --branch v2.99-R3 https://github.com/intel/openvino-ai-plugins-gimp.git
     ```
     > __Notes__:
     > This is an example that will create a `GIMP` folder in `C:\Users\Public\`, and then download the package to `C:\Users\Public\GIMP`, you still can define where to place this package by yourself.
@@ -119,19 +118,7 @@ Please download [gimp-2.99.16-setup.exe](https://download.gimp.org/gimp/v2.99/wi
 
     This step will take time for downloading necessary packages.
 
-
-- After creating python environment, gimpenv3, and downloaded necessary packages, you will be prompted to download models 
-- Enter __\<number\>__ for downloading models you want to run. You can specify multiple models separated by spaces:
-
-    ![](figs/model_downloding_SD1.5.png)
-
- - Download "__1__", SD-1.5 Square (512x512) as an instance. Once installation process is completed when you see the messages below.
-
-    ![](figs/download_SD1.5.png)
-
-    >**Notes:**
-    > - The downloaded models include FP16 and INT8 precision, and INT8 precision can be executed on MTL NPU.
-    > - Weights is saved at `C:\Users\<user_name>\openvino-ai-plugins-gimp\weights`.
+- After creating python environment, gimpenv3, and downloaded necessary packages, you can now start GIMP application from Start Menu. 
 
 # Execute Stable-Diffusion in GIMP
 
@@ -143,18 +130,36 @@ With previous section, the GIMP AI plugins with OpenVINO™ Backend is installed
 
 You can now simply open Gimp from the start menu as you would any normal Windows application.
 
-## Execute Stable-Diffusion – SD1.5_square_int8
+## Download & Install Stable Diffusion Models
+- Once GIMP application is started, Click "__Layer__" \> "__OpenVINO-AI-Plugins__" \> "__Stable diffusion__".
+- Once Stable Diffusion Plugin UI opens up, click on Model Button on the top left corner as show in the below picture:
 
-Please follow below steps to execute Stable-Diffusion - SD1.5_square_int8. For other features, please refer to [OpenVINO™ Image Generator Plugin with Stable Diffusion](https://github.com/intel/openvino-ai-plugins-gimp/tree/main?tab=readme-ov-file#openvino-image-generator-plugin-with-stable-diffusion) section.
+    ![](figs/stable-diffusion_model1.png)
+- This now opens the Stable Diffusion Model Management window from where you can install the required models by clicking on "Install" button next to the model name. 
 
-- Following previous section to launch GIMP
+    ![](figs/stable-diffusion_model2.png)  
 
-    ![](figs/gimp_launch.png)
+ - Download Stable Diffusion 1.5 Square as an instance. 
 
-- Click "__File__" \> "__New__" \> "__OK__" to new a layer in GIMP
+    ![](figs/stable-diffusion_model3.png)
+
+ - Once installation process is completed, you will see the "Install" button is updated to "Installed" and you will find the models in the drop-down list
 
-    ![](figs/gimp_create_image.png)
-    ![](figs/gimp_ok.png)
+    ![](figs/stable-diffusion_model4.png)
+
+    >**Notes:**
+    > - The downloaded models include FP16 and INT8 precision
+    > - Weights is saved at `C:\Users\<user_name>\openvino-ai-plugins-gimp\weights`.
+
+
+
+## Execute Stable-Diffusion – Stable Diffusion 1.5 INT8
+
+Please follow below steps to execute Stable-Diffusion - Stable Diffusion 1.5 INT8. For other features, please refer to [OpenVINO™ Image Generator Plugin with Stable Diffusion](https://github.com/intel/openvino-ai-plugins-gimp/tree/main?tab=readme-ov-file#openvino-image-generator-plugin-with-stable-diffusion) section.
+
+- Following previous section to launch GIMP
+
+    ![](figs/gimp_launch.png)
 
 - Click "__Layer__" \> "__OpenVINO-AI-Plugins__" \> "__Stable diffusion__".
 
@@ -177,10 +182,3 @@ Please follow below steps to execute Stable-Diffusion - SD1.5_square_int8. For o
     >   - Balanced
     > - If you wish to generate more images in single run, please modify the Number of Images section.
 
-### Detail of Power Mode
-
-| Power mode | Execution detail |
-|----------:|:----------------|
-| Best Performance | Text Device:   GPU<br>Unet Device:   GPU<br>Unet-Neg Device:   GPU<br>VAE Device:  GPU | 
-| Best Power Efficiency | Text Device:   NPU<br>Unet Device:   NPU<br>Unet-Neg Device:   NPU<br>VAE Device:  GPU |
-| Balanced | Text Device:   NPU<br>Unet Device:   NPU<br>Unet-Neg Device:   GPU<br>VAE Device:  GPU |
diff --git a/README.md b/README.md
@@ -30,11 +30,10 @@ For detailed steps and tips please refer to [Linux Installation Guide](./Docs/li
 # OpenVINO™ Image Generator Plugin with Stable Diffusion
 #### Power Mode is enabled for int8 SD models and systems that has a "NPU" only 
 #### A. Prompt to Image
-1. Create or choose a layer
+1. You can start with a blank canvas or create/choose a layer
 2. Select Stable Diffusion from the drop down list in layers -> OpenVINO-AI-Plugins
-3. Choose the desired model and device from the drop down list.
-   - note that the [Latent Consistency Model](https://huggingface.co/SimianLuo/LCM_Dreamshaper_v7) is now supported. Choose `sd_1.5_square_lcm`
-4. Click on "Load Models" to compile & load the model on the selected device. Wait for it to complete. Please note that you need to perform this step only if you change the model or device or both. For any subsequent runs just click "Run Inference"
+3. Choose the desired Model and Power Mode from the drop down list.
+4. Click on "Load Models" to compile & load the model on the device. Wait for it to complete. Please note that you need to perform this step only if you change the model or device or both. For any subsequent runs just click "Run Inference"
 5. Enter prompt and other parameters
 6. Click on “Generate”. Wait for the total inference steps to get completed.
 
@@ -81,13 +80,7 @@ For detailed steps and tips please refer to [Linux Installation Guide](./Docs/li
 6. Click on “Generate”. Wait for the total inference steps to get completed.
 
 
-![](gifs/stable-diffusion.png)
-
-_With Advanced Setting_
-
-![](gifs/stable-diffusion1.png)
-
-_With Power Mode in Advanced Setting for SD int8 based models_
+_With Advanced Setting and Power Mode_
 
 ![](gifs/stable-diffusion2.png)
 

diff --git a/gifs/stable-diffusion2.png b/gifs/stable-diffusion2.png
diff --git a/gimpopenvino/complete_install.py b/gimpopenvino/complete_install.py
@@ -13,10 +13,21 @@
 import gimpopenvino
 import platform
 import subprocess
+import shutil
+from pathlib import Path
 from openvino.runtime import Core
 
+def install_base_models(base_model_dir, repo_weights_dir):
+    for folder in os.scandir(repo_weights_dir):
+        model = os.path.basename(folder)
+        model_path = os.path.join(base_model_dir, model)
+        if not os.path.isdir(model_path):
+            print("Copying {} to {}".format(model, base_model_dir))
+            shutil.copytree(Path(folder), model_path)
 
-def setup_python_weights(install_location=None):
+    print("Setup done for base models")
+
+def setup_python_weights(install_location=None, repo_weights_dir=None):
     if not install_location:
         install_location = os.path.join(os.path.expanduser("~"), "openvino-ai-plugins-gimp")
 
@@ -30,6 +41,9 @@ def setup_python_weights(install_location=None):
     if not os.path.isdir(weight_path):
         os.mkdir(weight_path)
 
+    if repo_weights_dir:
+        install_base_models(weight_path, repo_weights_dir)
+
     step = 1
     print("\n##########\n")
     plugin_loc = os.path.dirname(gimpopenvino.__file__)