diff --git a/filters/f_decoder_wrapper.c b/filters/f_decoder_wrapper.c
index 3253eea95b..397c3a0a98 100644
--- a/filters/f_decoder_wrapper.c
+++ b/filters/f_decoder_wrapper.c
@@ -226,6 +226,7 @@ struct priv {
     char *cur_hwdec;
     char *decoder_desc;
     bool try_spdif;
+    bool attached_picture;
     bool pts_reset;
     int attempt_framedrops; // try dropping this many frames
     int dropped_frames; // total frames _probably_ dropped
@@ -524,6 +525,14 @@ void mp_decoder_wrapper_set_spdif_flag(struct mp_decoder_wrapper *d, bool spdif)
     pthread_mutex_unlock(&p->cache_lock);
 }
 
+void mp_decoder_wrapper_set_coverart_flag(struct mp_decoder_wrapper *d, bool c)
+{
+    struct priv *p = d->f->priv;
+    pthread_mutex_lock(&p->cache_lock);
+    p->attached_picture = c;
+    pthread_mutex_unlock(&p->cache_lock);
+}
+
 bool mp_decoder_wrapper_get_pts_reset(struct mp_decoder_wrapper *d)
 {
     struct priv *p = d->f->priv;
@@ -974,13 +983,9 @@ static void read_frame(struct priv *p)
     if (!frame.type)
         return;
 
-    if (p->header->attached_picture && frame.type == MP_FRAME_VIDEO) {
-        p->decoded_coverart = frame;
-        mp_filter_internal_mark_progress(p->decf);
-        return;
-    }
-
     pthread_mutex_lock(&p->cache_lock);
+    if (p->attached_picture && frame.type == MP_FRAME_VIDEO)
+        p->decoded_coverart = frame;
     if (p->attempt_framedrops) {
         int dropped = MPMAX(0, p->packets_without_output - 1);
         p->attempt_framedrops = MPMAX(0, p->attempt_framedrops - dropped);
@@ -988,6 +993,11 @@ static void read_frame(struct priv *p)
     }
     pthread_mutex_unlock(&p->cache_lock);
 
+    if (p->decoded_coverart.type) {
+        mp_filter_internal_mark_progress(p->decf);
+        return;
+    }
+
     p->packets_without_output = 0;
 
     if (p->preroll_discard && frame.type != MP_FRAME_EOF) {
diff --git a/filters/f_decoder_wrapper.h b/filters/f_decoder_wrapper.h
index e6f7705b33..9fa1a4f010 100644
--- a/filters/f_decoder_wrapper.h
+++ b/filters/f_decoder_wrapper.h
@@ -55,6 +55,9 @@ double mp_decoder_wrapper_get_container_fps(struct mp_decoder_wrapper *d);
 // Whether to prefer spdif wrapper over real decoders on next reinit.
 void mp_decoder_wrapper_set_spdif_flag(struct mp_decoder_wrapper *d, bool spdif);
 
+// Whether to decode only 1 frame and then stop, and cache the frame across resets.
+void mp_decoder_wrapper_set_coverart_flag(struct mp_decoder_wrapper *d, bool c);
+
 // True if a pts reset was observed (audio only, heuristic).
 bool mp_decoder_wrapper_get_pts_reset(struct mp_decoder_wrapper *d);
 
diff --git a/player/video.c b/player/video.c
index 2b8f03630e..c13bdb4074 100644
--- a/player/video.c
+++ b/player/video.c
@@ -261,9 +261,12 @@ void reinit_video_chain_src(struct MPContext *mpctx, struct track *track)
         vo_c->dec_src = track->dec->f->pins[0];
         vo_c->filter->container_fps =
             mp_decoder_wrapper_get_container_fps(track->dec);
-        vo_c->is_coverart = !!track->stream->attached_picture;
+        vo_c->is_coverart = !!track->attached_picture;
         vo_c->is_sparse = track->stream->still_image || vo_c->is_coverart;
 
+        if (vo_c->is_coverart)
+            mp_decoder_wrapper_set_coverart_flag(track->dec, true);
+
         track->vo_c = vo_c;
         vo_c->track = track;