init

.gitignore

@@ -0,0 +1,3 @@
+__pycache__
+
+/venv

README.md

@@ -0,0 +1,5 @@
+# Online Image Node
+
+A custom node for Comfyui that search for an image online with a seed input for reruns.
+
+Find in category 'Feur'.

__init__.py

@@ -0,0 +1,14 @@
+import sys
+import pathlib
+
+
+project_dir = pathlib.Path(__file__).parent.resolve()
+src_dir = project_dir / "src"
+sys.path.insert(1, str(src_dir))
+
+
+from online_image_node import (
+    NODE_CLASS_MAPPINGS,
+    NODE_DISPLAY_NAME_MAPPINGS,
+    OnlineImageNode
+)

requirements.txt

@@ -0,0 +1,3 @@
+duckduckgo_search==5.2.1
+numpy==1.24.2
+Pillow==10.3.0

setup.sh

@@ -0,0 +1,10 @@
+#!/bin/sh
+set -e
+cd "$(dirname "$(realpath "$0")")"
+
+
+python -m venv venv
+. venv/bin/activate
+
+
+pip install -r requirements.txt

src/online_image_node/__init__.py

@@ -0,0 +1,12 @@
+from .node import OnlineImageNode
+
+
+NODE_CLASS_MAPPINGS = {
+    "Online Image Search": OnlineImageNode
+}
+
+NODE_DISPLAY_NAME_MAPPINGS = {
+    "Online Image Search": "Online Image Search"
+}
+
+# WEB_DIRECTORY = "./web"

src/online_image_node/node.py

@@ -0,0 +1,97 @@
+from dataclasses import dataclass
+from pathlib import Path
+
+import numpy
+import torch
+from duckduckgo_search import DDGS
+from fastdownload import download_url
+from PIL import Image, ImageOps
+
+
+@dataclass
+class Query:
+    phrase: str
+    seed: int
+    pass
+
+
+class OnlineImageNode:
+    CATEGORY = "Feur"
+
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "search_phrase": (
+                    "STRING",
+                    {
+                        "default": "Feur",
+                        "multiline": True
+                    }
+                ),
+                "seed": (
+                    "NUMBER",
+                    {
+                        "default": 19
+                    }
+                )
+            }
+        }
+
+    OUTPUT_NODE = True
+    RETURN_TYPES = ["IMAGE"]
+    RETURN_NAMES = ["image"]
+
+
+    FUNCTION = "run"
+    def run(self, search_phrase: str, seed: int):
+        query = Query(search_phrase, seed)
+        path = self.download(query)
+        (image, _mask) = self.load_image(path)
+        return [image]
+
+
+    def download(self, query: Query):
+        with DDGS() as ddg:
+            results: list = ddg.images(query.phrase, max_results=50)
+            result = results[query.seed % len(results)]
+            filename = f"{query.phrase}_{query.seed}{Path(result['image']).suffix}"
+            dl_path = Path("/tmp") / filename
+            download_url(result["image"], dl_path)
+            return dl_path
+
+
+    def load_image(self, img_path: Path | str) -> tuple[torch.Tensor, torch.Tensor]:
+        """This code is taken from the LoadImage node code in ComfyUI. Maybe it's better to call that function directly?"""
+        img = Image.open(img_path)
+
+        # If the image has exif data, rotate it to the correct orientation and remove the exif data.
+        img_raw = ImageOps.exif_transpose(img)
+        assert img_raw is not None
+
+        # If in 32-bit mode, normalize the image appropriately.
+        if img_raw.mode == "I":
+            img_raw = img.point(lambda i: i * (1 / 255))
+
+        # If image is rgba, create mask.
+        if "A" in img_raw.getbands():
+            mask = numpy.array(img_raw.getchannel("A")).astype(numpy.float32) / 255.0
+            mask = 1.0 - torch.from_numpy(mask)
+        else:
+            # otherwise create a blank mask.
+            mask = torch.zeros(
+                (img_raw.height, img_raw.width), dtype=torch.float32, device="cpu"
+            )
+        mask = mask.unsqueeze(0)  # Add a batch dimension to mask
+
+        # Convert the image to RGB.
+        rgb_image = img_raw.convert("RGB")
+        # Normalize the image's rgb values to {x | x ∈ float32, 0 <= x <= 1}
+        rgb_image = numpy.array(rgb_image).astype(numpy.float32) / 255.0
+        # Convert the image to a tensor (torch.from_numpy gives a tensor with the format of [H, W, C])
+        rgb_image = torch.from_numpy(rgb_image)[
+            None,
+        ]  # Add a batch dimension, new format is [B, H, W, C]
+
+        return (rgb_image, mask)