Update README.md

README.md

@@ -1,6 +1,5 @@
-
 ---
-title: FakeSpotter Heuristic Demo
+title: FakeSpotter — Image Deepfake Detector
 emoji: 🔍
 colorFrom: blue
 colorTo: pink
@@ -10,20 +9,28 @@ app_file: app.py
 pinned: false
 ---
 
-# FakeSpotter (Heuristic, No-API, Free)
+## FakeSpotter — Image Deepfake Detector (Images Only)
+
+**What it does**
+- Detect faces (MTCNN). If faces are found, classify **each face**; otherwise classify the **whole image**.
+- Uses a vision classifier fine-tuned for deepfake detection from Hugging Face.
+- **No EXIF heuristics** — purely image-based.
 
-**Goal:** Upload an image → return a probability and reasons for **AI vs Real** using simple, free heuristics.
+**Model**
+- Default: `prithivMLmods/Deep-Fake-Detector-v2-Model`  
+  You can swap to another HF model by editing `MODEL_ID` in `app.py`.  
+  (Make sure the license allows your intended use.)
 
-## What it does
-- Read EXIF and look for camera fields (Make/Model) or generator keywords (e.g., Stable Diffusion, Midjourney).
-- Run **ELA (Error Level Analysis)** by recompressing at JPEG 95 and computing mean/std of differences.
-- Combine signals into a **0..1 AI-likelihood** score with a friendly label (Likely AI / Uncertain / Likely Real).
+**How to run on Spaces**
+1. Create a Space → SDK: **Gradio**, Template: **Blank**, Public.
+2. Upload `app.py`, `requirements.txt`, `README.md` to the **repo root**.
+3. Wait for “Running 🟢” and open the **App** tab.
 
-## How to deploy (Hugging Face Spaces)
-1. Create a new **Space** (SDK: Gradio, Template: Blank, Public).
-2. Upload `app.py`, `requirements.txt`, and this `README.md` to the root directory.
-3. Wait for the build → open **App** tab.
+**Notes**
+- This is a **classroom demo**, not a forensic tool.
+- For better robustness: constrain input size (done), and aggregate per-face scores (done).
+- For videos, extend by sampling frames and aggregating scores.
 
-## Disclaimer
-This is a **classroom demo**. Heuristics can be fooled by re-encoding, screenshots, or camera apps that strip EXIF.
-Do **not** rely on it for legal/forensic decisions.
+**License**
+- Your code: choose MIT/Apache-2.0 for simplicity.
+- The model’s own license is defined on its HF model card — verify before production use.