diff --git a/web/apps/photos/src/services/embeddingService.ts b/web/apps/photos/src/services/embeddingService.ts
index 04c5089f9e..f78d410042 100644
--- a/web/apps/photos/src/services/embeddingService.ts
+++ b/web/apps/photos/src/services/embeddingService.ts
@@ -1,3 +1,4 @@
+import type { EmbeddingModel } from "@/new/photos/services/embedding";
 import { inWorker } from "@/next/env";
 import log from "@/next/log";
 import { apiOrigin } from "@/next/origins";
@@ -9,7 +10,6 @@ import localForage from "@ente/shared/storage/localForage";
 import { getToken } from "@ente/shared/storage/localStorage/helpers";
 import type {
     Embedding,
-    EmbeddingModel,
     EncryptedEmbedding,
     GetEmbeddingDiffResponse,
     PutEmbeddingRequest,
diff --git a/web/apps/photos/src/types/embedding.tsx b/web/apps/photos/src/types/embedding.tsx
index 161244c159..c14b4a137d 100644
--- a/web/apps/photos/src/types/embedding.tsx
+++ b/web/apps/photos/src/types/embedding.tsx
@@ -1,11 +1,4 @@
-/**
- * The embeddings that we (the current client) knows how to handle.
- *
- * This is an exhaustive set of values we pass when PUT-ting encrypted
- * embeddings on the server. However, we should be prepared to receive an
- * {@link EncryptedEmbedding} with a model value different from these.
- */
-export type EmbeddingModel = "onnx-clip" | "file-ml-clip-face";
+import type { EmbeddingModel } from "@/new/photos/services/embedding";
 
 export interface EncryptedEmbedding {
     fileID: number;
diff --git a/web/packages/new/photos/services/embedding.ts b/web/packages/new/photos/services/embedding.ts
new file mode 100644
index 0000000000..9acd7ccd20
--- /dev/null
+++ b/web/packages/new/photos/services/embedding.ts
@@ -0,0 +1,46 @@
+/**
+ * The embeddings that we (the current client) knows how to handle.
+ *
+ * It is vernacularly called a model, but strictly speaking this is not the
+ * model, but the embedding produced by a particular model with a particular set
+ * of pre-/post- processing steps and related hyperparameters. It is better
+ * thought of as an "type" of embedding produced or consumed by the client.
+ *
+ * The embeddings themselves have a version included in them, so it is possible
+ * for us to make backward compatible updates to the indexing process on newer
+ * clients. Alternatively, if the changes are not backward compatible and can
+ * only be consumed by clients with the relevant scaffolding, then we change
+ * this "model" (i.e "type") field to create a new universe of embeddings.
+ *
+ * This is an exhaustive set of values we pass when GET-ing or PUT-ting
+ * encrypted embeddings from remote. However, we should be prepared to receive
+ * an {@link EncryptedEmbedding} with a model value different from these.
+ */
+export type EmbeddingModel = "onnx-clip" | "file-ml-clip-face";
+
+/**
+ * Ask remote for what all changes have happened to the face embeddings that it
+ * knows about since the last time we synced. Then update our local state to
+ * reflect those changes.
+ *
+ * It takes no parameters since it saves the last sync time in local storage.
+ */
+export const syncRemoteFaceEmbeddings = async () => {
+    return 0;
+};
+
+// const getFaceEmbeddings = async () => {
+
+// }
+
+/**
+ * GET /embeddings/diff for the given model and changes {@link sinceTime}.
+ *
+ * @param model The {@link EmbeddingModel} whose diff we wish for.
+ *
+ * @param sinceTime The last time we synced (epoch ms).
+ */
+// const getEmbeddingsDiff = async (model: EmbeddingModel, sinceTime: number) => {
+//     // const params = new URLSearchParams()
+
+// }