diff --git a/api/src/production/server.py b/api/src/production/server.py
index 88b7023..ce40939 100644
--- a/api/src/production/server.py
+++ b/api/src/production/server.py
@@ -45,7 +45,8 @@ MAX_FRAME_BYTES = 1 * 1024 * 1024
 MAX_IMAGE_PIXELS = 20_000_000
 MAX_WEBSOCKETS = 3
 MAX_WEBSOCKETS_PER_IP = 1
-FACE_MIN_INTERVAL = 1 / 2.1
+FACE_MAX_FPS = float(os.getenv("FACE_MAX_FPS", "5.5"))  # per websocket connection
+FACE_MIN_INTERVAL = 1 / FACE_MAX_FPS
 DECODE_DRAFT_SIZE = (256, 256)  # JPEG decodes at reduced scale, still larger than the 64px model input
 
 torch.set_num_threads(TORCH_THREADS)
diff --git a/website/src/Object_Detection.jsx b/website/src/Object_Detection.jsx
index c8f1383..63c6761 100644
--- a/website/src/Object_Detection.jsx
+++ b/website/src/Object_Detection.jsx
@@ -3,6 +3,11 @@ import { useRef, useEffect, useState } from 'react'
 
 const FRAME_QUALITY = 0.7;
 
+// How many frames per second are sent to the API. The server drops frames
+// arriving faster than ~2.1/s (FACE_MAX_FPS in server.py), so values
+// above that have no effect unless the server limit is raised too.
+const FRAMES_PER_SECOND = 5;
+
 function Object_Detection() {
     const videoRef = useRef(null);
     const canvasRefBBox = useRef(null);
@@ -125,7 +130,7 @@ function Object_Detection() {
             if (!canvas) return;
 
             ctx.clearRect(0, 0, canvas.width, canvas.height);
-            sendImageIntervall.current = setInterval(sendImage, 1000 / 2);
+            sendImageIntervall.current = setInterval(sendImage, 1000 / FRAMES_PER_SECOND);
         }
         socket.onmessage = (event) => {
             drawBBox(JSON.parse(event.data).bboxes);