diff --git a/README.md b/README.md
index f5337d2..fb6917f 100644
--- a/README.md
+++ b/README.md
@@ -61,3 +61,10 @@ Install from webui's Extensions tab.
 
 ## Precautions
 This program uses code that contains the Apache License 2.0
+
+# API
+Currently, there are two APIs available:
+ - To get the currently available SAM model: `GET http://localhost:7861/pbrem/sam-model`
+ - To process an image: `POST http://localhost:7861/pbrem/predict`
+
+After launching the web UI API, you can visit the detailed API documentation at `http://localhost:7861/redoc`
diff --git a/scripts/api.py b/scripts/api.py
new file mode 100644
index 0000000..ccdf13d
--- /dev/null
+++ b/scripts/api.py
@@ -0,0 +1,71 @@
+from typing import List
+
+import gradio as gr
+from PIL import Image
+from fastapi import FastAPI
+from pydantic import BaseModel, Field
+
+from modules.api.api import encode_pil_to_base64, decode_base64_to_image
+from scripts.main import process_image, model_list
+
+
+class PBRRemRequest(BaseModel):
+    img: str = Field(..., description='Image to process, base64 encoded')
+
+    td_abg_enabled: bool = False
+    h_split: int = 256
+    v_split: int = 256
+    n_cluster: int = 500
+    alpha: int = 50
+    th_rate: float = 0.1
+
+    cascadePSP_enabled: bool = False
+    fast: bool = False
+    psp_L: int = 900
+
+    sa_enabled: bool = False
+    model_name: str = Field('sam_vit_h_4b8939.pth',
+                            description='SAM model name')
+    query: str = Field('', description='segmentation prompt')
+    predicted_iou_threshold: float = 0.9
+    stability_score_threshold: float = 0.9
+    clip_threshold: float = 0.1
+
+
+class PBRRemResponse(BaseModel):
+    img: str
+    mask: str
+
+
+def pbrem_api(_: gr.Blocks, app: FastAPI):
+
+    @app.get("/pbrem/sam-model", description='query available SAM model', response_model=List[str])
+    async def get_pbrem_sam_model() -> List[str]:
+        return model_list
+
+    @app.post('/pbrem/predict', description='process image', response_model=PBRRemResponse)
+    async def post_pbrem_predict(payload: PBRRemRequest) -> PBRRemResponse:
+        print(f"PBRemTools API /pbrem/predict received request")
+        input_image = decode_base64_to_image(payload.img)
+        output_image, mask = process_image(
+            input_image,
+            payload.td_abg_enabled, payload.h_split, payload.v_split, payload.n_cluster, payload.alpha, payload.th_rate,
+            payload.cascadePSP_enabled, payload.fast, payload.psp_L,
+            payload.sa_enabled, payload.query, payload.model_name, payload.predicted_iou_threshold, payload.stability_score_threshold, payload.clip_threshold
+        )
+        base64_img = encode_pil_to_base64(Image.fromarray(output_image))
+        base64_mask = encode_pil_to_base64(Image.fromarray(mask))
+        # Compatible: low version webui will return base64 string directly
+        if not isinstance(base64_img, str):
+            base64_img = base64_img.decode()
+        if not isinstance(base64_mask, str):
+            base64_mask = base64_mask.decode()
+        return PBRRemResponse(img=base64_img, mask=base64_mask)
+
+
+try:
+    import modules.script_callbacks as script_callbacks
+    script_callbacks.on_app_started(pbrem_api)
+except Exception as e:
+    print(e)
+    print("PBRemTools API failed to initialize")