deepface scanning offloaded to modal cloud function

app.py

@@ -4,19 +4,30 @@ from detector.voice import verify_voices
 from detector.video import verify_faces_in_video
 from reports.pdf_report import generate_pdf_report
 from utils.youtube_utils import download_youtube_video
+import modal
+verify_faces_remote = modal.Function.lookup("deepface-agent", "verify_faces_remote")
+
 
 # Holds latest results
 last_face_result = None
 last_voice_result = None
 last_video_results = None
 
-
-def start_scan(image, audio):
-    return "Scanning in progress...", None
-
-def compare_faces(img1_path, img2_path):
+# @app.local_entrypoint()
+def compare_faces(img1_path: str, img2_path: str) -> str:
+    """Use this tool to compare to faces for a match
+    Args:
+        img1_path: The path to the first image
+        img2_path: The path to the second image
+    """
     global last_face_result
-    result = verify_faces(img1_path, img2_path)
+    
+    # Read image files as bytes
+    with open(img1_path, "rb") as f1, open(img2_path, "rb") as f2:
+        img1_bytes = f1.read()
+        img2_bytes = f2.read()
+        
+    result = verify_faces_remote.remote(img1_bytes, img2_bytes)
     result_text = ""
 
     if "error" in result:
@@ -26,14 +37,19 @@ def compare_faces(img1_path, img2_path):
         result_text = f"✅ Match! Distance: {result['distance']:.4f} (Threshold: {result['threshold']})"
         last_face_result = result_text
         return result_text
-    
+
     else:
         result_text = f"❌ No Match. Distance: {result['distance']:.4f} (Threshold: {result['threshold']})"
         last_face_result = result_text
         return result_text
 
 
-def compare_voices(audio1, audio2):
+def compare_voices(audio1: str, audio2: str) -> str:
+    """Use this tool to compare two voices for a match
+    Args:
+        audio1: The path to the first audio file
+        audio2: The path to the second audio file
+    """
     global last_voice_result
     result = verify_voices(audio1, audio2)
     result_text = ""
@@ -51,46 +67,35 @@ def compare_voices(audio1, audio2):
         return result_text
 
 
-def scan_video(video_file, ref_img, youtube_url=""):
+def scan_video(video_file: str, ref_img: str, youtube_url="") -> str:
+    """Use this tool to scan a video for deepfake face swaps
+    Args:
+        video_file: The path to the video file
+        ref_img: The path to the reference image
+        youtube_url: The YouTube URL (optional)
+    """
     global last_video_results
-    
+
     if youtube_url:
         try:
             video_file = download_youtube_video(youtube_url)
         except Exception as e:
             return f"❌ Error downloading YouTube video: {str(e)}"
-    
+
     results = verify_faces_in_video(video_file, ref_img)
     report = ""
-    last_video_results = results 
+    last_video_results = results
     for r in results:
         if "error" in r:
             report += f"\n⚠️ Frame {r['frame']}: {r['error']}"
-            # last_video_results.append(report)
+
         else:
             status = "✅ Match" if r["verified"] else "❌ Mismatch"
             report += f"\n🖼 Frame {r['frame']}: {status} (Distance: {r['distance']})"
 
-            # last_video_results.append(report)
     return report
 
 
-# def scan_video(video_path, ref_img):
-#     global last_video_results
-#     results = verify_faces_in_video(video_path, ref_img)
-#     report = ""
-#     last_video_results = results 
-#     for r in results:
-#         if "error" in r:
-#             report += f"\n⚠️ Frame {r['frame']}: {r['error']}"
-#             # last_video_results.append(report)
-#         else:
-#             status = "✅ Match" if r["verified"] else "❌ Mismatch"
-#             report += f"\n🖼 Frame {r['frame']}: {status} (Distance: {r['distance']})"
-
-#             # last_video_results.append(report)
-#     return report
-
 def generate_report():
     return generate_pdf_report(last_face_result, last_voice_result, last_video_results)
 
@@ -111,9 +116,7 @@ with gr.Blocks(title="Deepfake Watchdog") as demo:
         output_text = gr.Textbox(label="Result")
         # output_gallery = gr.Gallery(label="Matched Results")
 
-        run_button.click(
-            compare_faces, inputs=[image1, image2], outputs=[output_text]
-        )
+        run_button.click(compare_faces, inputs=[image1, image2], outputs=[output_text])
 
     # Voice Verification
     with gr.Tab("🎤 Voice Verification"):
@@ -126,30 +129,21 @@ with gr.Blocks(title="Deepfake Watchdog") as demo:
         voice_output = gr.Textbox(label="Result")
 
         voice_btn.click(compare_voices, inputs=[audio1, audio2], outputs=voice_output)
-        
-    # Video DeepFake Scan
-    # gr.Markdown("### 📹 Video Deepfake Scan")
-    # with gr.Tab("📹 Video Deepfake Scan"):
-    #     gr.Markdown("Upload a video and a reference image. We'll scan for deepfake face mismatches.")
-
-    #     ref_img = gr.Image(type="filepath", label="Reference Face")
-    #     video_input = gr.Video(label="Video File")
-    #     scan_btn = gr.Button("Scan Video")
-    #     scan_output = gr.Textbox(label="Scan Results", lines=10)
 
-    #     scan_btn.click(scan_video, inputs=[video_input, ref_img], outputs=scan_output)
-    
     with gr.Tab("📹 Video Deepfake Scan"):
-        gr.Markdown("🔍 Upload a video or paste a YouTube link and we'll analyze it for deepfake face swaps.")
-        
+        gr.Markdown(
+            "🔍 Upload a video or paste a YouTube link and we'll analyze it for deepfake face swaps."
+        )
+
         ref_img = gr.Image(type="filepath", label="Reference Face")
         video_input = gr.Video(label="Video File (optional)")
         youtube_url = gr.Textbox(label="YouTube URL (optional)")
         scan_btn = gr.Button("Scan Video")
         scan_output = gr.Textbox(label="Scan Results", lines=10)
 
-        scan_btn.click(scan_video, inputs=[video_input, ref_img, youtube_url], outputs=scan_output)    
-
+        scan_btn.click(
+            scan_video, inputs=[video_input, ref_img, youtube_url], outputs=scan_output
+        )
 
     with gr.Tab("📄 Generate Report"):
         report_btn = gr.Button("Generate PDF Report")

detector/face.py

@@ -1,6 +1,19 @@
 from deepface import DeepFace
 import cv2
 import os
+from modal_app.modal_app import verify_faces_remote
+from fastapi import UploadFile
+
+def face_verify_tool(img1: UploadFile, img2: UploadFile):
+    img1_bytes = img1.file.read()
+    img2_bytes = img2.file.read()
+    
+    results = verify_faces_remote(img1_bytes, img2_bytes)
+
+    return results
+
+
+
 
 def verify_faces(img1_path, img2_path, model_name="Facenet", detector_backend="opencv"):
     """

pyproject.toml

@@ -8,8 +8,10 @@ dependencies = [
     "deepface>=0.0.93",
     "fpdf>=1.7.2",
     "gradio[mcp]>=5.32.1",
+    "modal>=1.0.3",
     "opencv-python-headless>=4.11.0.86",
     "pydub>=0.25.1",
     "resemblyzer>=0.1.4",
     "tf-keras>=2.19.0",
+    "yt-dlp>=2025.5.22",
 ]