Stable-Diffusion-API-Server

This project is a RESTful API server that provides image generation and editing services based on Stable Diffusion models. The APIs are compatible with OpenAI APIs of image generation and editing.

[!NOTE] The project is still under active development. The existing features still need to be improved and more features will be added in the future.

Quick Start

Setup

Install WasmEdge v0.14.1

curl -sSf https://raw.githubusercontent.com/WasmEdge/WasmEdge/master/utils/install_v2.sh | bash -s -- -v 0.14.1

Deply wasmedge_stablediffusion plugin

For the purpose of demonstration, we will use the stable diffusion plugin for Mac Apple Silicon. You can find the plugin for other platforms Releases/0.14.1

# Download stable diffusion plugin for Mac Apple Silicon
curl -LO https://github.com/WasmEdge/WasmEdge/releases/download/0.14.1/WasmEdge-plugin-wasmedge_stablediffusion-0.14.1-darwin_arm64.tar.gz

# Unzip the plugin to $HOME/.wasmedge/plugin
tar -xzf WasmEdge-plugin-wasmedge_stablediffusion-0.14.1-darwin_arm64.tar.gz -C $HOME/.wasmedge/plugin

rm $HOME/.wasmedge/plugin/libwasmedgePluginWasiNN.dylib

Run sd-api-server

Download the stable diffusion model
```
curl -LO https://huggingface.co/second-state/stable-diffusion-v-1-4-GGUF/resolve/main/stable-diffusion-v1-4-Q8_0.gguf
```
The available stable diffusion models:

Download sd-api-server.wasm

curl -LO https://github.com/LlamaEdge/sd-api-server/releases/latest/download/sd-api-server.wasm

Start the server
```
wasmedge --dir .:. sd-api-server.wasm --model-name sd-v1.4 --model stable-diffusion-v1-4-Q8_0.gguf
```
[!TIP] sd-api-server will use 8080 port by default. You can change the port by adding --port <port>.
- Reduce the memory usage
In the default setting, the server support two tasks: text2image for image generations and image2image for image edits. If you want to run one of them, you can specify the task type by adding --task <task-type>. For example, if you only want to run image generations, then just start the server with the following command:
```
wasmedge --dir .:. sd-api-server.wasm --model-name sd-v1.4 --model stable-diffusion-v1-4-Q8_0.gguf --task text2image
```

Usage

Image Generation

Send a request for image generation

curl -X POST 'http://localhost:8080/v1/images/generations' \
--header 'Content-Type: application/json' \
--data '{
    "model": "sd-v1.4",
    "prompt": "A cute baby sea otter"
}'

If the request is handled successfully, the server will return a JSON response like the following:

{
"created": 1723431133,
"data": [
    {
        "url": "/archives/file_74f514a2-8d33-4f9d-bcc0-42e8db14ecbc/output.png",
        "prompt": "A cute baby sea otter"
    }
]
}

Preview the generated image

Image Editing

Send a request for image editing

curl --location 'http://localhost:8080/v1/images/edits' \
--form 'image=@"otter.png"' \
--form 'prompt="A cute baby sea otter with blue eyes"'

If the request is handled successfully, the server will return a JSON response like the following:

{
"created": 1723432689,
"data": [
    {
        "url": "/archives/file_554e4d53-6072-4988-83e6-fe684655a734/output.png",
        "prompt": "A cute baby sea otter with blue eyes"
    }
]
}

Preview the edited image

Build

For Linux users
```
cargo build --release
```

For macOS users

Download the wasi-sdk from the official website and unzip it to the directory you want.
Build the project

export WASI_SDK_PATH=/path/to/wasi-sdk
export CC="${WASI_SDK_PATH}/bin/clang --sysroot=${WASI_SDK_PATH}/share/wasi-sysroot"
cargo clean
cargo update
cargo build --release

If the build process is successful, sd-api-server.wasm will be generated in target/wasm32-wasip1/release/.

CLI Options

$ wasmedge target/wasm32-wasip1/release/sd-api-server.wasm -h

LlamaEdge-Stable-Diffusion API Server

Usage: sd-api-server.wasm [OPTIONS] --model-name <MODEL_NAME> <--model <MODEL>|--diffusion-model <DIFFUSION_MODEL>>

Options:
  -m, --model-name <MODEL_NAME>
          Sets the model name
      --model <MODEL>
          Path to full model [default: ]
      --diffusion-model <DIFFUSION_MODEL>
          Path to the standalone diffusion model file [default: ]
      --vae <VAE>
          Path to vae [default: ]
      --clip-l <CLIP_L>
          Path to the clip-l text encoder [default: ]
      --t5xxl <T5XXL>
          Path to the the t5xxl text encoder [default: ]
      --lora-model-dir <LORA_MODEL_DIR>
          Path to the lora model directory
      --control-net <CONTROL_NET>
          Path to control net model
      --control-net-cpu
          Keep controlnet on cpu (for low vram)
      --threads <THREADS>
          Number of threads to use during computation. Default is -1, which means to use all available threads [default: -1]
      --clip-on-cpu
          Keep clip on cpu (for low vram)
      --vae-on-cpu
          Keep vae on cpu (for low vram)
      --task <TASK>
          Task type [default: full] [possible values: text2image, image2image, full]
      --socket-addr <SOCKET_ADDR>
          Socket address of LlamaEdge API Server instance. For example, `0.0.0.0:8080`
      --port <PORT>
          Port number [default: 8080]
  -h, --help
          Print help (see more with '--help')
  -V, --version
          Print version

LlamaEdge / sd-api-server

readme