diff --git a/proposals/2529-text-messages-as-captions.md b/proposals/2529-text-messages-as-captions.md
new file mode 100644
index 00000000..df76e325
--- /dev/null
+++ b/proposals/2529-text-messages-as-captions.md
@@ -0,0 +1,37 @@
+# Use existing m.room.message/m.text events as captions for images
+
+## Background
+
+There is a demand to be able to apply a text caption to an image, as is possible in other chat platforms. In Matrix this is not possible, so people will generally send two events: one `m.image`, then a `m.text` event immediately afterward to simulate a caption.
+
+Better would be to able to explicitly mark an event as a caption.
+
+## Proposal
+
+Allow an optional `m.relates_to` field in the `content` field of a text message event.
+
+Example:
+
+```
+...
+  "content": {
+    "body": "Caption text",
+    "msgtype": "m.text",
+    "m.relates_to": {
+            "event_id": "$(some image event)",
+            "rel_type": "m.caption"
+        }
+  },
+```
+
+If a client recognises the `rel_type`, they can render the caption with the image rather than as a separate message in the timeline.
+
+The benefit of this is that if a client doesn't support or recognise the `m.caption`, it can ignore the relation and just render the message inline.
+
+This would not require aggregation from the server since there will always be a need to send the event separately anyway.
+
+## Potential issues
+
+* Not sure how this relates to the broader questions discussed in MSC1849
+* This is catering to a narrow use-case requirement. There may be a more general solution available
+* Would MSC1767 (extensible events) obsolete this?