diff --git a/desktop/src/main/services/ml.ts b/desktop/src/main/services/ml.ts
index 05642fc95f..cc1ae5764c 100644
--- a/desktop/src/main/services/ml.ts
+++ b/desktop/src/main/services/ml.ts
@@ -11,8 +11,11 @@ import { app, utilityProcess } from "electron/main";
 import path from "node:path";
 import log from "../log";
 
+/** The active ML worker (utility) process, if any. */
+let _child: UtilityProcess | undefined;
+
 /**
- * Create a new ML worker process.
+ * Create a new ML worker process, terminating the older ones (if any).
  *
  * [Note: ML IPC]
  *
@@ -68,6 +71,12 @@ import log from "../log";
  * to be relayed using `postMessage`.
  */
 export const createMLWorker = (window: BrowserWindow) => {
+    if (_child) {
+        log.debug(() => "Terminating previous ML worker process");
+        _child.kill();
+        _child = undefined;
+    }
+
     const { port1, port2 } = new MessageChannelMain();
 
     const child = utilityProcess.fork(path.join(__dirname, "ml-worker.js"));
@@ -77,6 +86,8 @@ export const createMLWorker = (window: BrowserWindow) => {
     window.webContents.postMessage("createMLWorker/port", undefined, [port2]);
 
     handleMessagesFromUtilityProcess(child);
+
+    _child = child;
 };
 
 /**
diff --git a/web/packages/base/types/ipc.ts b/web/packages/base/types/ipc.ts
index 748490ed58..c0644760c0 100644
--- a/web/packages/base/types/ipc.ts
+++ b/web/packages/base/types/ipc.ts
@@ -335,7 +335,7 @@ export interface Electron {
     // - ML
 
     /**
-     * Create a new ML worker.
+     * Create a new ML worker, terminating the older ones (if any).
      *
      * This creates a new Node.js utility process, and sets things up so that we
      * can communicate directly with that utility process using a
diff --git a/web/packages/new/photos/services/ml/index.ts b/web/packages/new/photos/services/ml/index.ts
index ba8083ad74..5b57dade21 100644
--- a/web/packages/new/photos/services/ml/index.ts
+++ b/web/packages/new/photos/services/ml/index.ts
@@ -98,22 +98,27 @@ export const terminateMLWorker = async () => {
  * Obtain a port from the Node.js layer that can be used to communicate with the
  * ML worker process.
  */
-const createMLWorker = async (electron: Electron): Promise<MessagePort> => {
-    electron.createMLWorker();
-
+const createMLWorker = (electron: Electron): Promise<MessagePort> => {
     // The main process will do its thing, and send back the port it created to
     // us by sending an message on the "createMLWorker/port" channel via the
     // postMessage API. This roundabout way is needed because MessagePorts
     // cannot be transferred via the usual send/invoke pattern.
 
-    return new Promise((resolve) => {
-        window.onmessage = ({ source, data, ports }: MessageEvent) => {
-            // The source check verifies that the message is coming from the
+    const port = new Promise<MessagePort>((resolve) => {
+        const l = ({ source, data, ports }: MessageEvent) => {
+            // The source check verifies that the message is coming from our own
             // preload script. The data is the message that was posted.
-            if (source == window && data == "createMLWorker/port")
+            if (source == window && data == "createMLWorker/port") {
+                window.removeEventListener("message", l);
                 resolve(ensure(ports[0]));
+            }
         };
+        window.addEventListener("message", l);
     });
+
+    electron.createMLWorker();
+
+    return port;
 };
 
 /**