diff --git a/filters/f_hwtransfer.c b/filters/f_hwtransfer.c
index 649d28497c..9640f406ad 100644
--- a/filters/f_hwtransfer.c
+++ b/filters/f_hwtransfer.c
@@ -35,6 +35,10 @@ struct priv {
     int *fmt_upload_index;
     int *fmt_upload_num;
 
+    // List of source formats that require hwmap instead of hwupload.
+    int *map_fmts;
+    int num_map_fmts;
+
     struct mp_hwupload public;
 };
 
@@ -58,6 +62,25 @@ static const struct ffmpeg_and_other_bugs shitlist[] = {
     {0}
 };
 
+struct hwmap_pairs {
+    int first_fmt;
+    int second_fmt;
+};
+
+// We cannot discover which pairs of hardware formats need to use hwmap to
+// convert between the formats, so we need a lookup table.
+static const struct hwmap_pairs hwmap_pairs[] = {
+    {
+        .first_fmt = IMGFMT_VAAPI,
+        .second_fmt = IMGFMT_VULKAN,
+    },
+    {
+        .first_fmt = IMGFMT_DRMPRIME,
+        .second_fmt = IMGFMT_VAAPI,
+    },
+    {0}
+};
+
 static bool select_format(struct priv *p, int input_fmt,
                           int *out_sw_fmt, int *out_upload_fmt)
 {
@@ -183,7 +206,19 @@ static void process(struct mp_filter *f)
         goto error;
     }
 
-    struct mp_image *dst = mp_av_pool_image_hw_upload(p->hw_pool, src);
+    struct mp_image *dst;
+    bool map_images = false;
+    for (int n = 0; n < p->num_map_fmts; n++) {
+        if (src->imgfmt == p->map_fmts[n]) {
+            map_images = true;
+            break;
+        }
+    }
+
+    if (map_images)
+        dst = mp_av_pool_image_hw_map(p->hw_pool, src);
+    else
+        dst = mp_av_pool_image_hw_upload(p->hw_pool, src);
     if (!dst)
         goto error;
 
@@ -296,6 +331,16 @@ static bool probe_formats(struct mp_hwupload *u, int hw_imgfmt)
         }
     }
 
+    for (int n = 0; hwmap_pairs[n].first_fmt; n++) {
+        if (hwmap_pairs[n].first_fmt == hw_imgfmt) {
+            MP_TARRAY_APPEND(p, p->map_fmts, p->num_map_fmts,
+                             hwmap_pairs[n].second_fmt);
+        } else if (hwmap_pairs[n].second_fmt == hw_imgfmt) {
+            MP_TARRAY_APPEND(p, p->map_fmts, p->num_map_fmts,
+                             hwmap_pairs[n].first_fmt);
+        }
+    }
+
     for (int n = 0; cstr->valid_sw_formats &&
                     cstr->valid_sw_formats[n] != AV_PIX_FMT_NONE; n++)
     {
diff --git a/filters/f_lavfi.c b/filters/f_lavfi.c
index cbb3066430..41a3ea4eac 100644
--- a/filters/f_lavfi.c
+++ b/filters/f_lavfi.c
@@ -561,8 +561,8 @@ static void init_graph(struct lavfi *c)
                 hwdec_ctx = hwdec_devices_get_first(info->hwdec_devs);
             }
             if (hwdec_ctx && hwdec_ctx->av_device_ref) {
-                MP_VERBOSE(c, "Configuring hwdec_interop=%s for filters\n",
-                           hwdec_ctx->driver_name);
+                MP_VERBOSE(c, "Configuring hwdec_interop=%s for filter graph: %s\n",
+                           hwdec_ctx->driver_name, c->graph_string);
                 for (int n = 0; n < c->graph->nb_filters; n++) {
                     AVFilterContext *filter = c->graph->filters[n];
                     filter->hw_device_ctx =
diff --git a/test/ref/img_formats.txt b/test/ref/img_formats.txt
index 2b1cca1522..cd84963b6a 100644
--- a/test/ref/img_formats.txt
+++ b/test/ref/img_formats.txt
@@ -1239,7 +1239,7 @@ videotoolbox: ctype=unknown
     planes=0, chroma=0:0 align=1:1
     {}
   AVD: name=videotoolbox_vld chroma=0:0 flags=0x8 [hw]
-vulkan: [GENERIC] ctype=unknown
+vulkan: ctype=unknown
   Basic desc: [le][be][hw]
     planes=0, chroma=0:0 align=1:1
     {}
diff --git a/video/fmt-conversion.c b/video/fmt-conversion.c
index 5194221d50..d200a5db31 100644
--- a/video/fmt-conversion.c
+++ b/video/fmt-conversion.c
@@ -67,6 +67,7 @@ static const struct {
     {IMGFMT_CUDA, AV_PIX_FMT_CUDA},
     {IMGFMT_P010, AV_PIX_FMT_P010},
     {IMGFMT_DRMPRIME, AV_PIX_FMT_DRM_PRIME},
+    {IMGFMT_VULKAN, AV_PIX_FMT_VULKAN},
 
     {0, AV_PIX_FMT_NONE}
 };
diff --git a/video/img_format.h b/video/img_format.h
index ebdbd66e29..7559c8c172 100644
--- a/video/img_format.h
+++ b/video/img_format.h
@@ -319,6 +319,7 @@ enum mp_imgfmt {
     IMGFMT_VDPAU_OUTPUT,    // VdpOutputSurface
     IMGFMT_VAAPI,
     IMGFMT_VIDEOTOOLBOX,    // CVPixelBufferRef
+    IMGFMT_VULKAN,          // VKImage
 
     // Generic pass-through of AV_PIX_FMT_*. Used for formats which don't have
     // a corresponding IMGFMT_ value.
diff --git a/video/mp_image_pool.c b/video/mp_image_pool.c
index 2a4a8940c0..d80c65d3b5 100644
--- a/video/mp_image_pool.c
+++ b/video/mp_image_pool.c
@@ -426,3 +426,30 @@ struct mp_image *mp_av_pool_image_hw_upload(struct AVBufferRef *hw_frames_ctx,
     mp_image_copy_attributes(dst, src);
     return dst;
 }
+
+struct mp_image *mp_av_pool_image_hw_map(struct AVBufferRef *hw_frames_ctx,
+                                         struct mp_image *src)
+{
+    AVFrame *dst_frame = av_frame_alloc();
+    if (!dst_frame)
+        return NULL;
+
+    dst_frame->format = ((AVHWFramesContext*)hw_frames_ctx->data)->format;
+    dst_frame->hw_frames_ctx = av_buffer_ref(hw_frames_ctx);
+
+    AVFrame *src_frame = mp_image_to_av_frame(src);
+    if (av_hwframe_map(dst_frame, src_frame, 0) < 0) {
+        av_frame_free(&src_frame);
+        av_frame_free(&dst_frame);
+        return NULL;
+    }
+    av_frame_free(&src_frame);
+
+    struct mp_image *dst = mp_image_from_av_frame(dst_frame);
+    av_frame_free(&dst_frame);
+    if (!dst)
+        return NULL;
+
+    mp_image_copy_attributes(dst, src);
+    return dst;
+}
diff --git a/video/mp_image_pool.h b/video/mp_image_pool.h
index 3c7e6221d1..ab9065e6e9 100644
--- a/video/mp_image_pool.h
+++ b/video/mp_image_pool.h
@@ -41,4 +41,6 @@ bool mp_update_av_hw_frames_pool(struct AVBufferRef **hw_frames_ctx,
 struct mp_image *mp_av_pool_image_hw_upload(struct AVBufferRef *hw_frames_ctx,
                                             struct mp_image *src);
 
+struct mp_image *mp_av_pool_image_hw_map(struct AVBufferRef *hw_frames_ctx,
+                                         struct mp_image *src);
 #endif