diff --git a/.github/workflows/build.yml b/.github/workflows/build.yml
index 739c4e17..ef97b47a 100644
--- a/.github/workflows/build.yml
+++ b/.github/workflows/build.yml
@@ -156,6 +156,7 @@ jobs:
         uses: docker/build-push-action@v5
         with:
           context: .
+          file: docker/Dockerfile
           platforms: |
             linux/amd64
             linux/386
@@ -215,7 +216,7 @@ jobs:
         uses: docker/build-push-action@v5
         with:
           context: .
-          file: hardware.Dockerfile
+          file: docker/hardware.Dockerfile
           platforms: linux/amd64
           push: ${{ github.event_name != 'pull_request' }}
           tags: ${{ steps.meta-hw.outputs.tags }}
diff --git a/.github/workflows/test.yml b/.github/workflows/test.yml
index f2089dec..5d9e7e25 100644
--- a/.github/workflows/test.yml
+++ b/.github/workflows/test.yml
@@ -79,6 +79,7 @@ jobs:
         uses: docker/build-push-action@v5
         with:
           context: .
+          file: docker/Dockerfile
           platforms: linux/${{ matrix.platform }}
           push: false
           load: true
@@ -92,7 +93,7 @@ jobs:
         uses: docker/build-push-action@v5
         with:
           context: .
-          file: hardware.Dockerfile
+          file: docker/hardware.Dockerfile
           platforms: linux/amd64
           push: false
           load: true
diff --git a/Dockerfile b/docker/Dockerfile
similarity index 79%
rename from Dockerfile
rename to docker/Dockerfile
index ba436825..34a96757 100644
--- a/Dockerfile
+++ b/docker/Dockerfile
@@ -5,16 +5,7 @@ ARG PYTHON_VERSION="3.11"
 ARG GO_VERSION="1.24"
 
 
-# 1. Download ngrok binary (for support arm/v6)
-FROM alpine AS ngrok
-ARG TARGETARCH
-ARG TARGETOS
-
-ADD https://bin.equinox.io/c/bNyj1mQVY4c/ngrok-v3-stable-${TARGETOS}-${TARGETARCH}.tgz /
-RUN tar -xzf /ngrok-v3-stable-${TARGETOS}-${TARGETARCH}.tgz -C /bin
-
-
-# 2. Build go2rtc binary
+# 1. Build go2rtc binary
 FROM --platform=$BUILDPLATFORM golang:${GO_VERSION}-alpine AS build
 ARG TARGETPLATFORM
 ARG TARGETOS
@@ -35,7 +26,7 @@ COPY . .
 RUN --mount=type=cache,target=/root/.cache/go-build CGO_ENABLED=0 go build -ldflags "-s -w" -trimpath
 
 
-# 3. Final image
+# 2. Final image
 FROM python:${PYTHON_VERSION}-alpine AS base
 
 # Install ffmpeg, tini (for signal handling),
@@ -55,7 +46,6 @@ RUN if [ "${TARGETARCH}" = "amd64" ]; then apk add --no-cache libva-intel-driver
 # RUN libva-vdpau-driver mesa-vdpau-gallium (+150MB total)
 
 COPY --from=build /build/go2rtc /usr/local/bin/
-COPY --from=ngrok /bin/ngrok /usr/local/bin/
 
 ENTRYPOINT ["/sbin/tini", "--"]
 VOLUME /config
diff --git a/hardware.Dockerfile b/docker/hardware.Dockerfile
similarity index 75%
rename from hardware.Dockerfile
rename to docker/hardware.Dockerfile
index e75a97cd..03b7d496 100644
--- a/hardware.Dockerfile
+++ b/docker/hardware.Dockerfile
@@ -5,15 +5,10 @@
 # https://packages.debian.org/trixie/ffmpeg
 ARG DEBIAN_VERSION="trixie-slim"
 ARG GO_VERSION="1.24-bookworm"
-ARG NGROK_VERSION="3"
-
-FROM debian:${DEBIAN_VERSION} AS base
-FROM golang:${GO_VERSION} AS go
-FROM ngrok/ngrok:${NGROK_VERSION} AS ngrok
 
 
 # 1. Build go2rtc binary
-FROM --platform=$BUILDPLATFORM go AS build
+FROM --platform=$BUILDPLATFORM golang:${GO_VERSION} AS build
 ARG TARGETPLATFORM
 ARG TARGETOS
 ARG TARGETARCH
@@ -31,34 +26,28 @@ COPY . .
 RUN --mount=type=cache,target=/root/.cache/go-build CGO_ENABLED=0 go build -ldflags "-s -w" -trimpath
 
 
-# 2. Collect all files
-FROM scratch AS rootfs
+# 2. Final image
+FROM debian:${DEBIAN_VERSION}
 
-COPY --link --from=build /build/go2rtc /usr/local/bin/
-COPY --link --from=ngrok /bin/ngrok /usr/local/bin/
-
-# 3. Final image
-FROM base
 # Prepare apt for buildkit cache
 RUN rm -f /etc/apt/apt.conf.d/docker-clean \
   && echo 'Binary::apt::APT::Keep-Downloaded-Packages "true";' >/etc/apt/apt.conf.d/keep-cache
-# Install ffmpeg, bash (for run.sh), tini (for signal handling),
+
+# Install ffmpeg, tini (for signal handling),
 # and other common tools for the echo source.
 # non-free for Intel QSV support (not used by go2rtc, just for tests)
 # mesa-va-drivers for AMD APU
 # libasound2-plugins for ALSA support
 RUN --mount=type=cache,target=/var/cache/apt,sharing=locked --mount=type=cache,target=/var/lib/apt,sharing=locked \
     echo 'deb http://deb.debian.org/debian trixie non-free' > /etc/apt/sources.list.d/debian-non-free.list && \
-    apt-get -y update && apt-get -y install tini ffmpeg \
+    apt-get -y update && apt-get -y install ffmpeg tini \
         python3 curl jq \
         intel-media-va-driver-non-free \
         mesa-va-drivers \
         libasound2-plugins && \
     apt-get clean && rm -rf /var/lib/apt/lists/*
 
-COPY --link --from=rootfs / /
-
-
+COPY --from=build /build/go2rtc /usr/local/bin/
 
 ENTRYPOINT ["/usr/bin/tini", "--"]
 VOLUME /config
diff --git a/internal/ffmpeg/ffmpeg.go b/internal/ffmpeg/ffmpeg.go
index 25d61e4b..8eba0a0b 100644
--- a/internal/ffmpeg/ffmpeg.go
+++ b/internal/ffmpeg/ffmpeg.go
@@ -129,8 +129,9 @@ var defaults = map[string]string{
 	// hardware Rockchip
 	// important to use custom ffmpeg https://github.com/AlexxIT/go2rtc/issues/768
 	// hevc - doesn't have a profile setting
-	"h264/rkmpp": "-c:v h264_rkmpp_encoder -g 50 -bf 0 -profile:v high -level:v 4.1",
-	"h265/rkmpp": "-c:v hevc_rkmpp_encoder -g 50 -bf 0 -level:v 5.1",
+	"h264/rkmpp":  "-c:v h264_rkmpp -g 50 -bf 0 -profile:v high -level:v 4.1",
+	"h265/rkmpp":  "-c:v hevc_rkmpp -g 50 -bf 0 -profile:v main -level:v 5.1",
+	"mjpeg/rkmpp": "-c:v mjpeg_rkmpp",
 
 	// hardware NVidia on Linux and Windows
 	// preset=p2 - faster, tune=ll - low latency
diff --git a/internal/ffmpeg/ffmpeg_test.go b/internal/ffmpeg/ffmpeg_test.go
index 2ab1170d..30052d78 100644
--- a/internal/ffmpeg/ffmpeg_test.go
+++ b/internal/ffmpeg/ffmpeg_test.go
@@ -251,15 +251,33 @@ func _TestParseArgsHwV4l2m2m(t *testing.T) {
 }
 
 func TestParseArgsHwRKMPP(t *testing.T) {
-	// [HTTP-MJPEG] video will be transcoded to H264
-	args := parseArgs("http://example.com#video=h264#hardware=rkmpp")
-	require.Equal(t, `ffmpeg -hide_banner -fflags nobuffer -flags low_delay -i http://example.com -c:v h264_rkmpp_encoder -g 50 -bf 0 -profile:v high -level:v 4.1 -an -user_agent ffmpeg/go2rtc -rtsp_transport tcp -f rtsp {output}`, args.String())
-
-	args = parseArgs("http://example.com#video=h264#rotate=180#hardware=rkmpp")
-	require.Equal(t, `ffmpeg -hide_banner -fflags nobuffer -flags low_delay -i http://example.com -c:v h264_rkmpp_encoder -g 50 -bf 0 -profile:v high -level:v 4.1 -an -vf "transpose=1,transpose=1" -user_agent ffmpeg/go2rtc -rtsp_transport tcp -f rtsp {output}`, args.String())
-
-	args = parseArgs("http://example.com#video=h264#height=320#hardware=rkmpp")
-	require.Equal(t, `ffmpeg -hide_banner -fflags nobuffer -flags low_delay -i http://example.com -c:v h264_rkmpp_encoder -g 50 -bf 0 -profile:v high -level:v 4.1 -height 320 -an -user_agent ffmpeg/go2rtc -rtsp_transport tcp -f rtsp {output}`, args.String())
+	tests := []struct {
+		name   string
+		source string
+		expect string
+	}{
+		{
+			name:   "[FILE] transcoding to H264",
+			source: "bbb.mp4#video=h264#hardware=rkmpp",
+			expect: `ffmpeg -hide_banner -hwaccel rkmpp -hwaccel_output_format drm_prime -afbc rga -re -i bbb.mp4 -c:v h264_rkmpp -g 50 -bf 0 -profile:v high -level:v 4.1 -an -user_agent ffmpeg/go2rtc -rtsp_transport tcp -f rtsp {output}`,
+		},
+		{
+			name:   "[FILE] transcoding with rotation",
+			source: "bbb.mp4#video=h264#rotate=180#hardware=rkmpp",
+			expect: `ffmpeg -hide_banner -hwaccel rkmpp -hwaccel_output_format drm_prime -afbc rga -re -i bbb.mp4 -c:v h264_rkmpp -g 50 -bf 0 -profile:v high -level:v 4.1 -an -vf "format=drm_prime|nv12,hwupload,vpp_rkrga=transpose=4" -user_agent ffmpeg/go2rtc -rtsp_transport tcp -f rtsp {output}`,
+		},
+		{
+			name:   "[FILE] transcoding with scaling",
+			source: "bbb.mp4#video=h264#height=320#hardware=rkmpp",
+			expect: `ffmpeg -hide_banner -hwaccel rkmpp -hwaccel_output_format drm_prime -afbc rga -re -i bbb.mp4 -c:v h264_rkmpp -g 50 -bf 0 -profile:v high -level:v 4.1 -an -vf "format=drm_prime|nv12,hwupload,scale_rkrga=-1:320:force_original_aspect_ratio=0" -user_agent ffmpeg/go2rtc -rtsp_transport tcp -f rtsp {output}`,
+		},
+	}
+	for _, test := range tests {
+		t.Run(test.name, func(t *testing.T) {
+			args := parseArgs(test.source)
+			require.Equal(t, test.expect, args.String())
+		})
+	}
 }
 
 func _TestParseArgsHwCuda(t *testing.T) {
diff --git a/internal/ffmpeg/hardware/README.md b/internal/ffmpeg/hardware/README.md
new file mode 100644
index 00000000..2d7f21cf
--- /dev/null
+++ b/internal/ffmpeg/hardware/README.md
@@ -0,0 +1,106 @@
+# Hardware
+
+You **DON'T** need hardware acceleration if:
+
+- you not using [FFmpeg source](https://github.com/AlexxIT/go2rtc#source-ffmpeg)
+- you using only `#video=copy` for FFmpeg source
+- you using only `#audio=...` (any audio) transcoding for FFmpeg source
+
+You **NEED** hardware acceleration if you using `#video=h264`, `#video=h265`, `#video=mjpeg` (video) transcoding.
+
+## Important
+
+- Acceleration is disabled by default because it can be unstable (it can be changed in future)
+- go2rtc can automatically detect supported hardware acceleration if enabled
+- go2rtc will enable hardware decoding only if hardware encoding supported
+- go2rtc will use the same GPU for decoder and encoder
+- Intel and AMD will switch to software decoder if input codec is not supported with hardware decoder
+- NVidia will fail if input codec is not supported with hardware decoder
+- Raspberry always uses software decoder
+
+```yaml
+streams:
+  # auto select hardware encoder
+  camera1_hw: ffmpeg:rtsp://rtsp:12345678@192.168.1.123/av_stream/ch0#video=h264#hardware
+  
+  # manual select hardware encoder (vaapi, cuda, v4l2m2m, dxva2, videotoolbox)
+  camera1_vaapi: ffmpeg:rtsp://rtsp:12345678@192.168.1.123/av_stream/ch0#video=h264#hardware=vaapi
+```
+
+## Docker and Hass Addon
+
+There are two versions of the Docker container and Hass Add-on:
+
+- Latest (alpine) support hardware acceleration for Intel iGPU (CPU with Graphics) and Raspberry.
+- Hardware (debian 12) support Intel iGPU, AMD GPU, NVidia GPU.
+
+## Intel iGPU
+
+**Supported on:** Windows binary, Linux binary, Docker, Hass Addon.
+
+If you have Intel CPU Sandy Bridge (2011) with Graphics, you already have support hardware decoding/encoding for `AVC/H.264`.
+
+If you have Intel CPU Skylake (2015) with Graphics, you already have support hardware decoding/encoding for `AVC/H.264`, `HEVC/H.265` and `MJPEG`.
+
+Read more [here](https://en.wikipedia.org/wiki/Intel_Quick_Sync_Video#Hardware_decoding_and_encoding) and [here](https://en.wikipedia.org/wiki/Intel_Graphics_Technology#Capabilities_(GPU_video_acceleration)).
+
+Linux and Docker:
+
+- It may be important to have the latest version of the OS with the latest version of the Linux kernel. For example, on my **Debian 10 (kernel 4.19)** it did not work, but after update to **Debian 11 (kernel 5.10)** all was fine.
+- In case of troube check you have `/dev/dri/` folder on your host.
+
+Docker users should add `--privileged` option to container for access to Hardware.
+
+**PS.** Supported via [VAAPI](https://trac.ffmpeg.org/wiki/Hardware/VAAPI) engine on Linux and [DXVA2+QSV](https://trac.ffmpeg.org/wiki/Hardware/QuickSync) engine on Windows.
+
+## AMD GPU
+
+*I don't have the hardware for test support!!!*
+
+**Supported on:** Linux binary, Docker, Hass Addon.
+
+Docker users should install: `alexxit/go2rtc:master-hardware`. Docker users should add `--privileged` option to container for access to Hardware.
+
+Hass Addon users should install **go2rtc master hardware** version.
+
+**PS.** Supported via [VAAPI](https://trac.ffmpeg.org/wiki/Hardware/VAAPI) engine.
+
+## NVidia GPU
+
+**Supported on:** Windows binary, Linux binary, Docker.
+
+Docker users should install: `alexxit/go2rtc:master-hardware`.
+
+Read more [here](https://docs.frigate.video/configuration/hardware_acceleration) and [here](https://jellyfin.org/docs/general/administration/hardware-acceleration/#nvidia-hardware-acceleration-on-docker-linux).
+
+**PS.** Supported via [CUDA](https://trac.ffmpeg.org/wiki/HWAccelIntro#CUDANVENCNVDEC) engine.
+
+## Raspberry Pi 3
+
+**Supported on:** Linux binary, Docker, Hass Addon.
+
+I don't recommend using transcoding on the Raspberry Pi 3. It's extreamly slow, even with hardware acceleration. Also it may fail when transcoding 2K+ stream.
+
+## Raspberry Pi 4
+
+*I don't have the hardware for test support!!!*
+
+**Supported on:** Linux binary, Docker, Hass Addon.
+
+**PS.** Supported via [v4l2m2m](https://lalitm.com/hw-encoding-raspi/) engine.
+
+## macOS
+
+In my tests, transcoding is faster on the M1 CPU than on the M1 GPU. Transcoding time on M1 CPU better than any Intel iGPU and comparable to NVidia RTX 2070.
+
+**PS.** Supported via [videotoolbox](https://trac.ffmpeg.org/wiki/HWAccelIntro#VideoToolbox) engine.
+
+## Rockchip
+
+- Important to use custom FFmpeg with Rockchip support from [@nyanmisaka](https://github.com/nyanmisaka/ffmpeg-rockchip)
+  - Static binaries from [@MarcA711](https://github.com/MarcA711/Rockchip-FFmpeg-Builds/releases/)
+- Important to have Linux kernel 5.10 or 6.1
+
+**Tested**
+
+- [Orange Pi 3B](https://www.armbian.com/orangepi3b/) with Armbian 6.1, support transcoding H264, H265, MJPEG
diff --git a/internal/ffmpeg/hardware/hardware.go b/internal/ffmpeg/hardware/hardware.go
index 39ce3323..80166890 100644
--- a/internal/ffmpeg/hardware/hardware.go
+++ b/internal/ffmpeg/hardware/hardware.go
@@ -128,19 +128,32 @@ func MakeHardware(args *ffmpeg.Args, engine string, defaults map[string]string)
 		case EngineRKMPP:
 			args.Codecs[i] = defaults[name+"/"+engine]
 
-			for j, filter := range args.Filters {
-				if strings.HasPrefix(filter, "scale=") {
-					args.Filters = append(args.Filters[:j], args.Filters[j+1:]...)
+			if !args.HasFilters("drawtext=") {
+				args.Input = "-hwaccel rkmpp -hwaccel_output_format drm_prime -afbc rga " + args.Input
 
-					width, height, _ := strings.Cut(filter[6:], ":")
-					if width != "-1" {
-						args.Codecs[i] += " -width " + width
+				for i, filter := range args.Filters {
+					if strings.HasPrefix(filter, "scale=") {
+						args.Filters[i] = "scale_rkrga=" + filter[6:] + ":force_original_aspect_ratio=0"
 					}
-					if height != "-1" {
-						args.Codecs[i] += " -height " + height
+					if strings.HasPrefix(filter, "transpose=") {
+						if filter == "transpose=1,transpose=1" { // 180 degrees half-turn
+							args.Filters[i] = "vpp_rkrga=transpose=4" // reversal
+						} else {
+							args.Filters[i] = "vpp_rkrga=transpose=" + filter[10:]
+						}
 					}
-					break
 				}
+
+				if len(args.Filters) > 0 {
+					// fix if input doesn't support hwaccel, do nothing when support
+					// insert as first filter before hardware scale and transpose
+					args.InsertFilter("format=drm_prime|nv12,hwupload")
+				}
+			} else {
+				// enable software pixel for drawtext, scale and transpose
+				args.Input = "-hwaccel rkmpp -hwaccel_output_format nv12 -afbc rga " + args.Input
+
+				args.AddFilter("hwupload")
 			}
 		}
 	}
diff --git a/internal/ffmpeg/hardware/hardware_unix.go b/internal/ffmpeg/hardware/hardware_unix.go
index 4f688ce4..e8000e17 100644
--- a/internal/ffmpeg/hardware/hardware_unix.go
+++ b/internal/ffmpeg/hardware/hardware_unix.go
@@ -11,8 +11,9 @@ import (
 const (
 	ProbeV4L2M2MH264 = "-f lavfi -i testsrc2 -t 1 -c h264_v4l2m2m -f null -"
 	ProbeV4L2M2MH265 = "-f lavfi -i testsrc2 -t 1 -c hevc_v4l2m2m -f null -"
-	ProbeRKMPPH264   = "-f lavfi -i testsrc2 -t 1 -c h264_rkmpp_encoder -f null -"
-	ProbeRKMPPH265   = "-f lavfi -i testsrc2 -t 1 -c hevc_rkmpp_encoder -f null -"
+	ProbeRKMPPH264   = "-f lavfi -i testsrc2 -t 1 -c h264_rkmpp -f null -"
+	ProbeRKMPPH265   = "-f lavfi -i testsrc2 -t 1 -c hevc_rkmpp -f null -"
+	ProbeRKMPPJPEG   = "-f lavfi -i testsrc2 -t 1 -c mjpeg_rkmpp -f null -"
 	ProbeVAAPIH264   = "-init_hw_device vaapi -f lavfi -i testsrc2 -t 1 -vf format=nv12,hwupload -c h264_vaapi -f null -"
 	ProbeVAAPIH265   = "-init_hw_device vaapi -f lavfi -i testsrc2 -t 1 -vf format=nv12,hwupload -c hevc_vaapi -f null -"
 	ProbeVAAPIJPEG   = "-init_hw_device vaapi -f lavfi -i testsrc2 -t 1 -vf format=nv12,hwupload -c mjpeg_vaapi -f null -"
@@ -39,6 +40,10 @@ func ProbeAll(bin string) []*api.Source {
 				Name: runToString(bin, ProbeRKMPPH265),
 				URL:  "ffmpeg:...#video=h265#hardware=" + EngineRKMPP,
 			},
+			{
+				Name: runToString(bin, ProbeRKMPPJPEG),
+				URL:  "ffmpeg:...#video=mjpeg#hardware=" + EngineRKMPP,
+			},
 		}
 	}
 
@@ -83,6 +88,10 @@ func ProbeHardware(bin, name string) string {
 			if run(bin, ProbeRKMPPH265) {
 				return EngineRKMPP
 			}
+		case "mjpeg":
+			if run(bin, ProbeRKMPPJPEG) {
+				return EngineRKMPP
+			}
 		}
 
 		return EngineSoftware