diff --git a/build/cross/Dockerfile.arm64 b/build/cross/Dockerfile.arm64
index dbf059b7..2aa20c2d 100644
--- a/build/cross/Dockerfile.arm64
+++ b/build/cross/Dockerfile.arm64
@@ -60,6 +60,7 @@ RUN dpkg --add-architecture arm64
 # Install cross-compiler and native build tools
 RUN apt-get update && apt-get install -y --no-install-recommends \
     build-essential \
+    patch \
     pkg-config \
     cmake \
     nasm \
@@ -250,6 +251,9 @@ RUN github_prefix="" \
     && make install \
     && rm -rf /tmp/opus /tmp/opus.tar.gz
 
+# Reject raw formats smaller than the requested frame before any V4L2 upload.
+COPY build/cross/patches/0001-v4l2-m2m-reject-undersized-encoder-format.patch /tmp/ffmpeg-v4l2-size.patch
+
 # Download and build FFmpeg with RKMPP support
 RUN mkdir -p /tmp/ffmpeg-build && cd /tmp/ffmpeg-build \
     && github_prefix="" \
@@ -261,6 +265,7 @@ RUN mkdir -p /tmp/ffmpeg-build && cd /tmp/ffmpeg-build \
     && git remote add origin "${github_prefix}https://github.com/nyanmisaka/ffmpeg-rockchip.git" \
     && git fetch --depth 1 origin ${FFMPEG_ROCKCHIP_REV} \
     && git checkout --detach FETCH_HEAD \
+    && patch -p1 < /tmp/ffmpeg-v4l2-size.patch \
     && cd .. \
     # Build RKMPP
     && mkdir -p rkmpp/build && cd rkmpp/build \
diff --git a/build/cross/Dockerfile.armv7 b/build/cross/Dockerfile.armv7
index b1c52ef5..3e1931e4 100644
--- a/build/cross/Dockerfile.armv7
+++ b/build/cross/Dockerfile.armv7
@@ -60,6 +60,7 @@ RUN dpkg --add-architecture armhf
 # Install cross-compiler and native build tools
 RUN apt-get update && apt-get install -y --no-install-recommends \
     build-essential \
+    patch \
     pkg-config \
     cmake \
     nasm \
@@ -239,6 +240,9 @@ RUN github_prefix="" \
     && make install \
     && rm -rf /tmp/opus /tmp/opus.tar.gz
 
+# Reject raw formats smaller than the requested frame before any V4L2 upload.
+COPY build/cross/patches/0001-v4l2-m2m-reject-undersized-encoder-format.patch /tmp/ffmpeg-v4l2-size.patch
+
 # Download and build FFmpeg with RKMPP support
 RUN mkdir -p /tmp/ffmpeg-build && cd /tmp/ffmpeg-build \
     && github_prefix="" \
@@ -250,6 +254,7 @@ RUN mkdir -p /tmp/ffmpeg-build && cd /tmp/ffmpeg-build \
     && git remote add origin "${github_prefix}https://github.com/nyanmisaka/ffmpeg-rockchip.git" \
     && git fetch --depth 1 origin ${FFMPEG_ROCKCHIP_REV} \
     && git checkout --detach FETCH_HEAD \
+    && patch -p1 < /tmp/ffmpeg-v4l2-size.patch \
     && cd .. \
     # Build RKMPP
     && mkdir -p rkmpp/build && cd rkmpp/build \
diff --git a/build/cross/patches/0001-v4l2-m2m-reject-undersized-encoder-format.patch b/build/cross/patches/0001-v4l2-m2m-reject-undersized-encoder-format.patch
new file mode 100644
index 00000000..1688b34b
--- /dev/null
+++ b/build/cross/patches/0001-v4l2-m2m-reject-undersized-encoder-format.patch
@@ -0,0 +1,36 @@
+--- a/libavcodec/v4l2_m2m_enc.c
++++ b/libavcodec/v4l2_m2m_enc.c
+@@ -338,6 +338,7 @@
+     V4L2m2mPriv *priv = avctx->priv_data;
+     enum AVPixelFormat pix_fmt_output;
+     uint32_t v4l2_fmt_output;
++    unsigned int output_width, output_height;
+     int ret;
+ 
+     ret = ff_v4l2_m2m_create_context(priv, &s);
+@@ -370,6 +371,25 @@
+         v4l2_fmt_output = output->format.fmt.pix_mp.pixelformat;
+     else
+         v4l2_fmt_output = output->format.fmt.pix.pixelformat;
++
++    /* Drivers may clamp the requested raw format to their supported size.
++     * The software-frame upload still uses avctx/frame dimensions, so a
++     * smaller negotiated format would overrun the mapped output buffers.
++     * Allow upward alignment, but reject frames that do not fit.
++     */
++    if (V4L2_TYPE_IS_MULTIPLANAR(output->type)) {
++        output_width  = output->format.fmt.pix_mp.width;
++        output_height = output->format.fmt.pix_mp.height;
++    } else {
++        output_width  = output->format.fmt.pix.width;
++        output_height = output->format.fmt.pix.height;
++    }
++    if (output_width < avctx->width || output_height < avctx->height) {
++        av_log(avctx, AV_LOG_ERROR,
++               "Encoder cannot fit requested %dx%d frame in negotiated %ux%u format\n",
++               avctx->width, avctx->height, output_width, output_height);
++        return AVERROR(EINVAL);
++    }
+ 
+     pix_fmt_output = ff_v4l2_format_v4l2_to_avfmt(v4l2_fmt_output, AV_CODEC_ID_RAWVIDEO);
+     if (pix_fmt_output != avctx->pix_fmt) {