notebooks/synthscope_quickstart.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# SynthScope quickstart\n",
    "\n",
    "[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/SMSD75/SynthScope/blob/main/notebooks/synthscope_quickstart.ipynb)\n",
    "\n",
    "Run state-of-the-art **AI-generated image & video detectors** in a few lines with the [`synthscope`](https://github.com/SMSD75/SynthScope) library.\n",
    "\n",
    "> Tip: switch the runtime to a **GPU** (Runtime \u2192 Change runtime type) \u2014 the backbones are large."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 1. Install\n",
    "\n",
    "Installs `synthscope` (which pulls PyTorch, transformers, etc.) and OpenAI CLIP (needed by a couple of detectors). Takes a few minutes on first run."
   ]
  },
  {
   "cell_type": "code",
   "metadata": {},
   "execution_count": null,
   "outputs": [],
   "source": [
    "%pip install -q \"git+https://github.com/SMSD75/SynthScope.git\" \"git+https://github.com/openai/CLIP.git\""
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 2. Get detector weights\n",
    "\n",
    "This notebook uses **CuRe**, whose authors publish their weights on GitHub (a 25 MB adapter). The first prediction also downloads the frozen PE-Core backbone (2.7 GB) from Hugging Face. Other detectors need their weights as `<key>.pth` in the same folder; each model's README links the official download."
   ]
  },
  {
   "cell_type": "code",
   "metadata": {},
   "execution_count": null,
   "outputs": [],
   "source": [
    "import os, urllib.request\n",
    "\n",
    "CKPT = \"/content/synthscope_ckpts\" if os.path.isdir(\"/content\") else os.path.expanduser(\"~/synthscope_ckpts\")\n",
    "os.makedirs(CKPT, exist_ok=True)\n",
    "os.environ[\"SYNTHSCOPE_CKPT_DIR\"] = CKPT          # must be set before importing synthscope\n",
    "\n",
    "if not os.path.exists(f\"{CKPT}/cure.pth\"):\n",
    "    urllib.request.urlretrieve(\n",
    "        \"https://raw.githubusercontent.com/manic-cui/CuRe/main/weights/cure_adapter.pt\", f\"{CKPT}/cure.pth\")\n",
    "\n",
    "import synthscope\n",
    "[m[\"key\"] for m in synthscope.list_models() if m[\"available\"]]   # detectors that can run here"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 3. Upload an image (or video)"
   ]
  },
  {
   "cell_type": "code",
   "metadata": {},
   "execution_count": null,
   "outputs": [],
   "source": [
    "try:\n",
    "    from google.colab import files\n",
    "    uploaded = files.upload()\n",
    "    path = next(iter(uploaded))\n",
    "except Exception:\n",
    "    path = \"your_image.png\"  # running outside Colab: set a local path\n",
    "print(\"file:\", path)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 4. Detect"
   ]
  },
  {
   "cell_type": "code",
   "metadata": {},
   "execution_count": null,
   "outputs": [],
   "source": [
    "det = synthscope.load(\"cure\")\n",
    "result = det.predict(path)\n",
    "print(f\"Verdict: {result['verdict'].upper()}  |  P(AI-generated) = {result['fake_probability']:.3f}\")\n",
    "result"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Next steps\n",
    "\n",
    "- Add more detectors: put their weights in the `CKPT` folder as `<key>.pth` (see each model's README), then `synthscope.load(key)`.\n",
    "- Training-free video detectors need no weights: `synthscope.load('d3').predict('clip.mp4')`.\n",
    "- Use your own trained checkpoint: `synthscope.load('cnn', checkpoint='/path/to.pth')`.\n",
    "- See the [repo](https://github.com/SMSD75/SynthScope) for the web demo, the benchmarks (datasets download themselves), and how to add a detector."
   ]
  }
 ],
 "metadata": {
  "accelerator": "GPU",
  "colab": {
   "provenance": []
  },
  "kernelspec": {
   "display_name": "Python 3",
   "name": "python3"
  },
  "language_info": {
   "name": "python"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 0
}