app.py

import gradio as gr
import torch
import numpy as np
import cv2
import matplotlib.pyplot as plt
from PIL import Image
from transformers import AutoImageProcessor, SiglipForImageClassification

# ✅ Load model from Hugging Face (no manual files)
model_name = "prithivMLmods/deepfake-detector-model-v1"
processor = AutoImageProcessor.from_pretrained(model_name)
model = SiglipForImageClassification.from_pretrained(model_name)
model.eval()

# ✅ Haar cascade for face detection
face_detector = cv2.CascadeClassifier(cv2.data.haarcascades + "haarcascade_frontalface_default.xml")

# ✅ Deepfake analysis logic
def detect_deepfake(video_path):
    if video_path is None:
        return "❌ Please upload a valid .mp4 video", None

    cap = cv2.VideoCapture(video_path)
    preds = []
    count = 0
    max_frames = 20

    while True:
        ret, frame = cap.read()
        if not ret or count >= max_frames:
            break

        gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
        faces = face_detector.detectMultiScale(gray, 1.1, 4)

        if len(faces) > 0:
            x, y, w, h = faces[0]  # Take first detected face
            face = frame[y:y+h, x:x+w]
            if face.size == 0:
                continue

            face_rgb = cv2.cvtColor(face, cv2.COLOR_BGR2RGB)
            pil_img = Image.fromarray(face_rgb)
            inputs = processor(images=pil_img, return_tensors="pt")

            with torch.no_grad():
                logits = model(**inputs).logits
                prob = torch.softmax(logits, dim=-1)[0][1].item()
                preds.append(prob)

        count += 1

    cap.release()

    if not preds:
        return "❌ No faces detected. Try a clearer video.", None

    avg_conf = np.mean(preds)
    label = "**FAKE**" if avg_conf > 0.5 else "**REAL**"
    result = f"""
    🎯 **Result:** {label}  
    🔢 Avg Confidence: {avg_conf:.2f}  
    📊 Frames Analyzed: {len(preds)}
    """

    # ✅ Create histogram
    fig, ax = plt.subplots()
    ax.hist(preds, bins=10, color="red" if avg_conf > 0.5 else "green", edgecolor="black")
    ax.set_title("Fake Confidence per Frame")
    ax.set_xlabel("Fake Probability (0 = Real, 1 = Fake)")
    ax.set_ylabel("Frames")
    ax.grid(True)

    return result, fig

# ✅ Gradio interface
with gr.Blocks() as demo:
    gr.Markdown("## 🎭 Deepfake Detector (Transformer-based)")
    gr.Markdown("Upload a short `.mp4` video. The app will detect faces and classify the video as **REAL** or **FAKE**.")

    video_input = gr.Video(label="📤 Upload your video")
    result_output = gr.Markdown()
    graph_output = gr.Plot()
    analyze_button = gr.Button("🔍 Analyze")

    analyze_button.click(fn=detect_deepfake, inputs=video_input, outputs=[result_output, graph_output])

demo.queue().launch()