DocsICP FilingConsole
Documentation
Home Page Platform For AI Use Cases Use cases for DSW Quickly launch Stable Diffusion WebUI using DSW

Quickly launch Stable Diffusion WebUI using DSW

Updated at:
Copy as MD
Product
My Favorites

This topic describes how to quickly launch Stable Diffusion WebUI in DSW to perform model inference with a Stable Diffusion (SD) model.

Background information

Artificial Intelligence Generated Content (AIGC) uses artificial intelligence (AI) technology to automatically create content. It is the next major industrial trend after the Internet era. Text-to-image generation is a popular cross-modality task that creates images from text descriptions. This tutorial shows an example of launching Stable-Diffusion-WebUI to perform SD model inference. The following figure shows the expected result.

Prepare the environment and resources

  1. Activate PAI and create a workspace. For more information, see Activate PAI and create a default workspace.

  2. Create a DSW instance. Configure the key parameters as follows. For more information, see Create a DSW instance.

    • Region and Zone: For this tutorial, select one of the following regions: China (Beijing), China (Shanghai), China (Hangzhou), or China (Shenzhen). These regions provide faster download speeds for the ChatGLM model data.

    • Instance Type: ecs.gn6v-c8g1.2xlarge.

    • Image: From Alibaba Cloud Image, select stable-diffusion-webui-develop:1.0-pytorch2.0-gpu-py310-cu117-ubuntu22.04.

Step 1: Download the stable-diffusion-webui open source library and other dependencies

  1. Open the DSW instance, go to the Launcher page, select the Notebook tab, and then click Python 3 under Notebook in the QuickStart area.

  2. In the Notebook, run the following code to download the files.

    import os

! apt update
! apt install -y aria2

def aria2(url, filename, d):
    !aria2c --console-log-level=error -c -x 16 -s 16 {url} -o {filename} -d {d}
    
url_prefix = {
    "cn-shanghai": "http://pai-vision-data-sh.oss-cn-shanghai-internal.aliyuncs.com",
    "cn-hangzhou": "http://pai-vision-data-hz2.oss-cn-hangzhou-internal.aliyuncs.com",
    "cn-shenzhen": "http://pai-vision-data-sz.oss-cn-shenzhen-internal.aliyuncs.com",
    "cn-beijing": "http://pai-vision-data-bj.oss-cn-beijing-internal.aliyuncs.com", 
}

dsw_region = os.environ.get("dsw_region")
prefix = url_prefix[dsw_region] if dsw_region in url_prefix else "http://pai-vision-data-sh.oss-cn-shanghai.aliyuncs.com"


webui_url = f"{prefix}/aigc-data/code/stable-diffusion-webui-v1.tar.gz"
aria2(webui_url, webui_url.split("/")[-1], "./")

Step 2: Install common plug-ins

This example installs the tagcomplete and Chinese localization plug-ins by default. You can also add other plug-ins as needed.

! tar -xf stable-diffusion-webui-v1.tar.gz
! cd stable-diffusion-webui && wget -c http://pai-vision-data-sh.oss-cn-shanghai.aliyuncs.com/aigc-data/webui_config/config.json

Step 3: Download the model

This example uses the open source SD model Counterfeit-v2.5 as the base model. To ensure a stable download, a cached link is provided on OSS. Run the following command to download the cached model.

model_url = f"{prefix}/aigc-data/sd_models/Counterfeit-V2.5_fp16.safetensors"
aria2(model_url, model_url.split("/")[-1], "stable-diffusion-webui/models/Stable-diffusion")

vae_url = f"{prefix}/aigc-data/vae_models/Counterfeit-V2.5.vae.pt"
aria2(vae_url, vae_url.split("/")[-1], "stable-diffusion-webui/models/VAE")

embedding_url = f"{prefix}/aigc-data/embedding/EasyNegative.safetensors"
aria2(embedding_url, embedding_url.split("/")[-1], "stable-diffusion-webui/embeddings")

clip_url = f"{prefix}/aigc-data/clip/ViT-L-14.pt"
! mkdir -p /root/.cache/clip
aria2(clip_url, clip_url.split("/")[-1], " /root/.cache/clip")

bert_url = f"{prefix}/aigc-data/hug_model/models--bert-base-uncased.tar.gz"
aria2(bert_url, bert_url.split("/")[-1], "~/.cache/huggingface/hub")
! cd ~/.cache/huggingface/hub && tar -xvf models--bert-base-uncased.tar.gz

Alternatively, you can download the models from their open source addresses:

After the models are downloaded, you can also download other types of SD models from ModelScope or HuggingFace. Save them to the ./stable-diffusion-webui/models directory. On the WebUI page, you can then switch to these models for inference.

Step 4: Launch the WebUI

  1. Run the following command to launch the WebUI.

    ! cd stable-diffusion-webui && python launch.py --no-half-vae --xformers

  2. In the output, click the URL that follows Running on public URL, such as http://127.0.0.1:7860, to open the WebUI page. You can then perform model inference on this page.

    Note

    Because http://127.0.0.1:7860 is an internal endpoint, you can access the WebUI page only by clicking the link from within the current DSW instance. You cannot access it directly from an external browser.

Step 5: Test the model

After you complete the preceding steps, the WebUI for the AIGC text-to-image model is deployed. You can now test the deployment by performing model inference on the WebUI page. Configure the parameters as follows, and then click Generate.

image.png

  • VAE model: Counterfeit-V2.5.vae.pt

  • Prompts:

    • Prompt

      ((masterpiece,best quality)),1girl, solo, animal ears, rabbit, barefoot, knees up, dress, sitting, rabbit ears, short sleeves, looking at viewer, grass, short hair, smile, white hair, puffy sleeves, outdoors, puffy short sleeves, bangs, on ground, full body, animal, white dress, sunlight, brown eyes, dappled sunlight, day, depth of field

    • Negative prompt (Used to prevent or reduce unwanted content in the generated image)

      EasyNegative, extra fingers,fewer fingers

  • Sampler: DPM++2M Karras

  • Hires. fix: Click image.png and configure the following parameters:

    • Denoising strength: 0.6

    • Upscale by: 1.8

  • Height: 832

  • CFG Scale: 10

  • Seed: 2337269170

References

Previous：Fine-tune LLaMA 3 with LLaMA Factory on DSWNext：Virtual try-on: Fine-tune a Stable Diffusion LoRA