diff --git a/commons/src/commonMain/kotlin/com/vitorpamplona/amethyst/commons/richtext/RichTextParser.kt b/commons/src/commonMain/kotlin/com/vitorpamplona/amethyst/commons/richtext/RichTextParser.kt
index b19a86f8c2..378b2e9145 100644
--- a/commons/src/commonMain/kotlin/com/vitorpamplona/amethyst/commons/richtext/RichTextParser.kt
+++ b/commons/src/commonMain/kotlin/com/vitorpamplona/amethyst/commons/richtext/RichTextParser.kt
@@ -60,9 +60,9 @@ class RichTextParser {
 
         val contentType = frags[MimeTypeTag.TAG_NAME] ?: tags[MimeTypeTag.TAG_NAME]?.firstOrNull()
 
-        val isImage: Boolean
-        val isVideo: Boolean
-        val isPdf: Boolean
+        var isImage = false
+        var isVideo = false
+        var isPdf = false
 
         if (contentType != null) {
             isImage = contentType.startsWith("image/")
@@ -77,7 +77,17 @@ class RichTextParser {
             isImage = fullUrl.startsWith("data:image/")
             isVideo = fullUrl.startsWith("data:video/") || fullUrl.startsWith("data:audio/")
             isPdf = fullUrl.startsWith("data:application/pdf")
-        } else {
+        }
+
+        // Fall back to file-extension detection when the type is still unknown. This covers both
+        // the no-MIME case and a *malformed* imeta MIME — e.g. Primal iOS emits `m jpeg` instead
+        // of `m image/jpeg`, which matches none of the `startsWith` prefixes above. Without this
+        // fallback such a URL returns null and drops to a plain link: that discards the imeta
+        // `dim`/blurhash (so the loading placeholder can't reserve the image's height and the
+        // feed jumps once the bitmap arrives) and forces a needless URL-preview network
+        // round-trip just to rediscover the type the imeta already declared. `data:` URIs carry
+        // their type in the prefix, so a miss there is genuine — don't extension-probe them.
+        if (!isImage && !isVideo && !isPdf && !fullUrl.startsWith("data:")) {
             val removedParamsFromUrl = removeQueryParamsForExtensionComparison(fullUrl)
             isImage = imageExtensions.any { removedParamsFromUrl.endsWith(it) }
             isVideo = videoExtensions.any { removedParamsFromUrl.endsWith(it) }
diff --git a/commons/src/commonTest/kotlin/com/vitorpamplona/amethyst/commons/richtext/PdfParserTest.kt b/commons/src/commonTest/kotlin/com/vitorpamplona/amethyst/commons/richtext/PdfParserTest.kt
index 8492017c4e..2dc2f906fa 100644
--- a/commons/src/commonTest/kotlin/com/vitorpamplona/amethyst/commons/richtext/PdfParserTest.kt
+++ b/commons/src/commonTest/kotlin/com/vitorpamplona/amethyst/commons/richtext/PdfParserTest.kt
@@ -66,4 +66,66 @@ class PdfParserTest {
         assertTrue(RichTextParser.isPdfUrl("https://example.com/doc.PDF"))
         assertTrue(RichTextParser.isPdfUrl("https://example.com/doc.pdf?sig=abc"))
     }
+
+    // Primal iOS writes a bare subtype (`m jpeg`) instead of a full MIME (`m image/jpeg`).
+    // The bare subtype matches none of the `image/`/`video/`/`application/pdf` prefixes, so before
+    // the extension fallback the whole imeta was dropped: the URL rendered as a plain link, losing
+    // the `dim` needed to reserve the image's height (feed jump) and forcing a URL-preview fetch.
+    // The `.jpg` extension must still route it to a MediaUrlImage that carries `dim`.
+    @Test
+    fun detectsImageFromMalformedImetaMimeWithImageExtension() {
+        val url = "https://blossom.primal.net/33e7c01afbea894a64e1db44dece460b09a2426108f47143754e1cf4bfdf747c.jpg"
+        val content = "\n$url"
+        val tags =
+            ImmutableListOfLists(
+                arrayOf(
+                    arrayOf("imeta", "url $url", "m jpeg", "dim 1009.0x680.0"),
+                    arrayOf("client", "Primal iOS"),
+                ),
+            )
+
+        val state = RichTextParser().parseText(content, tags, null)
+
+        val imageMedia = state.mediaForPager[url]
+        assertTrue(imageMedia is MediaUrlImage, "Expected MediaUrlImage despite the malformed `m jpeg` mime")
+        assertEquals("1009x680", imageMedia.dim?.toString(), "The imeta dim must survive so the loader can reserve space")
+        assertEquals(1009f / 680f, imageMedia.dim?.aspectRatio())
+
+        val segment = state.paragraphs.flatMap { it.words }.first { it.segmentText == url }
+        assertTrue(segment is ImageSegment, "Expected ImageSegment, got ${segment::class.simpleName}")
+    }
+
+    // A malformed video subtype (`m mp4`) must likewise fall back to the `.mp4` extension.
+    @Test
+    fun detectsVideoFromMalformedImetaMimeWithVideoExtension() {
+        val url = "https://cdn.example.com/clip.mp4"
+        val tags =
+            ImmutableListOfLists(
+                arrayOf(
+                    arrayOf("imeta", "url $url", "m mp4"),
+                ),
+            )
+
+        val state = RichTextParser().parseText(url, tags, null)
+
+        val videoMedia = state.mediaForPager[url]
+        assertTrue(videoMedia is MediaUrlVideo, "Expected MediaUrlVideo despite the malformed `m mp4` mime")
+    }
+
+    // A malformed mime on a URL with *no* recognizable extension can't be recovered — it stays a
+    // link. This documents the boundary of the fallback so it isn't mistaken for a regression.
+    @Test
+    fun malformedImetaMimeWithoutExtensionStaysUnclassified() {
+        val url = "https://files.example.com/abcd1234"
+        val tags =
+            ImmutableListOfLists(
+                arrayOf(
+                    arrayOf("imeta", "url $url", "m jpeg"),
+                ),
+            )
+
+        val state = RichTextParser().parseText(url, tags, null)
+
+        assertEquals(null, state.mediaForPager[url], "No extension to recover from -> not treated as media")
+    }
 }