blackmagic-io 0.17.0b5

Python library for video I/O with Blackmagic DeckLink devices

Blackmagic DeckLink Python I/O Library

Continuation of blackmagic-output (now archived) with input/capture support added; the package was renamed in 0.16.0b0 to reflect the broader scope.

A Python library for video I/O with Blackmagic DeckLink devices using the official DeckLink SDK. This library provides a simple interface for displaying static frames, solid colors, and dynamic content from NumPy arrays, as well as for capturing frames to NumPy arrays.

Written by Nick Shaw, www.antlerpost.com, with a lot of help from Claude Code!

⚠️ Note: The library has only had minimal testing at this time, and is under ongoing development. Please report any issues you encounter. I am particularly interested in feedback from Linux and Windows users.

Features

Output

• Static Frame Output: Display static images from NumPy arrays
• Solid Color Output: Display solid colors for testing and calibration
• Dynamic Updates: Update currently displayed frame
• Multiple Resolutions: Support for all display modes supported by your DeckLink device (SD, HD, 2K, 4K, 8K, and PC modes)
• 8 and 10-bit Y'CbCr 4:2:2 output (PixelFormat.YUV8, PixelFormat.YUV10; 10-bit is the default for uint16/float input)
• 8, 10 and 12-bit R'G'B' 4:4:4 output (PixelFormat.BGRA, PixelFormat.RGB10, PixelFormat.RGB12; 8-bit HDMI only)
• HDR Support: SMPTE ST 2086 / CEA-861.3 HDR static metadata
• Y'CbCr matrix control: Rec.601 (SD only), Rec.709 (HD+), and Rec.2020 (HD+) matrix support

Input

• Video Capture: Capture video frames from DeckLink devices
• Automatic Format Conversion: Convert all DeckLink pixel formats to R'G'B' float
• Format Detection: Automatic detection of input signal format (resolution, frame rate, colorspace, EOTF)
• Metadata Access: Access to format metadata (pixel format, colorspace, EOTF, source range)
• Timecode Capture: Automatic extraction of embedded timecode (RP188 VITC/LTC/HFRTC)
• HDMI EDID Configuration: Advertises SDR, HDR PQ, and HDR HLG support over HDMI by default so HDR sources transmit HDR Static Metadata (the SDK default omits HLG); the advertised bitmask is configurable via set_hdmi_input_dynamic_ranges()

General

• Cross-Platform: Works on Windows, macOS, and Linux (mainly tested on macOS; basic Windows testing has confirmed library output and pixel_reader build with hardware; Linux untested with hardware)

Requirements

System Requirements

• Python 3.8 or higher
• Blackmagic DeckLink device (DeckLink, UltraStudio, or Intensity series)
• Blackmagic Desktop Video software installed

Python Dependencies

Python dependencies (NumPy >= 1.19.0, pybind11 >= 2.6.0) are automatically installed (if needed) during the build process.

DeckLink SDK

SDK v14.1 headers for all platforms are included in the repository - no separate download needed.

Installation

1. DeckLink SDK

All Platforms (macOS, Windows, Linux):

• _vendor/decklink_sdk/Mac/include/ - macOS headers
• _vendor/decklink_sdk/Win/include/ - Windows headers
• _vendor/decklink_sdk/Linux/include/ - Linux headers

The build system (CMake + scikit-build-core) automatically uses the correct platform-specific headers.

⚠️ Important: This library was built against SDK v14.1 to maintain compatibility with older macOS versions. If you need to download the SDK separately, ensure you get v14.1 from the Blackmagic Design developer site. Newer versions (v15.0+) may cause API compatibility issues and build failures.

2. Install the Library

Option A (recommended for most users): from PyPI

pip install blackmagic-io

This installs the latest beta. pip normally skips pre-releases, but falls back to them when no stable version is available — and every published version of this library is currently a beta (0.17.0bN). Once a stable 0.17.0 (or later) is published, pip install blackmagic-io will resolve to that stable version, and getting a future beta will require pip install --pre blackmagic-io. Pre-built wheels are available for Python 3.8–3.14 on macOS, Linux, and Windows; pip falls back to a source build on unsupported Pythons (which requires a C++ compiler — Xcode Command Line Tools on macOS, build-essential on Linux, or MSVC on Windows).

To use the library at runtime you also need Blackmagic Desktop Video installed on your system (separate from this Python package) — the runtime DeckLink driver and framework are provided by Desktop Video, available from blackmagicdesign.com/support.

Option B (for contributors or for modifying the source): from a clone

# Clone the repository
git clone https://github.com/nick-shaw/blackmagic-io.git
cd blackmagic-io

# Switch to the development branch
# (main holds released versions; work in progress lives on dev)
git checkout dev

# Initialize submodules (required for the advanced T-Pat example only)
git submodule update --init --recursive

# Install in development mode (this also installs numpy and pybind11 dependencies)
pip install -e .

# If upgrading from a previous development version, force reinstall:
pip install --force-reinstall -e .

3. Install Optional Dependencies

For examples and additional functionality:

pip install opencv-python imageio pillow jsonschema

Note: While imageio / PIL can load 16-bit TIFF files correctly, 16-bit PNG files are often converted to 8-bit during loading due to PIL limitations. For reliable 16-bit workflows, use TIFF format.

Quick Start

Output Example

import numpy as np
from blackmagic_io import BlackmagicOutput, DisplayMode

# Create a simple test image (1080p R'G'B', normalized float)
frame = np.ones((1080, 1920, 3), dtype=np.float32)
frame[:, :] = [1.0, 0.0, 0.0]  # Red frame

# Display the frame
with BlackmagicOutput() as output:
    # Initialize device (uses first available device)
    output.initialize()

    # Display static frame at 1080p25
    output.display_static_frame(frame, DisplayMode.HD1080p25)

    # Keep displaying (Enter to stop)
    input("Press Enter to stop...")

Note: The explicit initialize() call is optional - display_static_frame() will automatically initialize the first available device (device_index=0) if not already initialized. Explicit initialization is useful for device selection, better error handling, and timing control:

# Simpler alternative - auto-initialization (uses first device)
with BlackmagicOutput() as output:
    output.display_static_frame(frame, DisplayMode.HD1080p25)
    input("Press Enter to stop...")

Input Example

from blackmagic_io import BlackmagicInput

# Capture a frame from DeckLink input
with BlackmagicInput() as input_device:
    # Initialize device (uses first available device)
    input_device.initialize()

    # Capture frame as R'G'B' float array (0.0-1.0 range)
    rgb_frame = input_device.capture_frame_as_rgb(timeout_ms=5000)

    if rgb_frame is not None:
        print(f"Captured frame: {rgb_frame.shape}, dtype: {rgb_frame.dtype}")
        print(f"Value range: {rgb_frame.min():.3f} - {rgb_frame.max():.3f}")
        # Process frame data...
    else:
        print("No signal or timeout")

Capture with metadata:

from blackmagic_io import BlackmagicInput

with BlackmagicInput() as input_device:
    input_device.initialize()

    # Capture frame with format information
    frame_data = input_device.capture_frame_with_metadata(timeout_ms=5000)

    if frame_data is not None:
        print(f"Resolution: {frame_data['width']}x{frame_data['height']}")
        print(f"Format: {frame_data['format']}")
        print(f"Colorspace: {frame_data['colorspace']}")
        print(f"EOTF: {frame_data['eotf']}")
        print(f"Narrow range source: {frame_data['input_narrow_range']}")

        # Access timecode if present
        if 'timecode' in frame_data:
            tc = frame_data['timecode']
            separator = ';' if tc['is_drop_frame'] else ':'
            print(f"Timecode: {tc['hours']:02d}:{tc['minutes']:02d}:{tc['seconds']:02d}{separator}{tc['frames']:02d}")

        # Access HDR metadata if present
        if 'hdr_metadata' in frame_data:
            hdr = frame_data['hdr_metadata']
            if 'display_primaries' in hdr:
                print(f"Display Primaries:")
                print(f"  Red:   ({hdr['display_primaries']['red']['x']:.4f}, "
                      f"{hdr['display_primaries']['red']['y']:.4f})")
                print(f"  Green: ({hdr['display_primaries']['green']['x']:.4f}, "
                      f"{hdr['display_primaries']['green']['y']:.4f})")
                print(f"  Blue:  ({hdr['display_primaries']['blue']['x']:.4f}, "
                      f"{hdr['display_primaries']['blue']['y']:.4f})")
            if 'mastering_luminance' in hdr:
                print(f"Mastering Luminance: {hdr['mastering_luminance']['max']:.1f} / "
                      f"{hdr['mastering_luminance']['min']:.4f} cd/m²")

        # Access R'G'B' data
        rgb = frame_data['rgb']  # float32 array (H×W×3)
        # Process frame...

API Reference

The library provides APIs for both output and input:

High-level Python wrappers:

• blackmagic_io.BlackmagicOutput - Convenient API for video output
• blackmagic_io.BlackmagicInput - Convenient API for video capture

Low-level direct access:

• decklink_io.DeckLinkOutput - Fine-grained control for output
• decklink_io.DeckLinkInput - Fine-grained control for capture

High-Level API: BlackmagicOutput Class

Convenient Python wrapper for most video output operations.

Methods

get_available_devices() -> List[str] Get list of available DeckLink device names.

get_device_capabilities(device_index=0) -> dict Get device capabilities (name and supported input/output).

• device_index: Index of device to query (default: 0)
• Returns: Dictionary with:
  • 'name': Device name
  • 'supports_input': True if device can capture video
  • 'supports_output': True if device can output video

Example:

from blackmagic_io import BlackmagicOutput

output = BlackmagicOutput()
caps = output.get_device_capabilities(0)
print(f"Device: {caps['name']}")
print(f"Supports input: {caps['supports_input']}")
print(f"Supports output: {caps['supports_output']}")

Note: Some DeckLink devices are single-direction only (e.g., UltraStudio Monitor 3G supports output only, UltraStudio Recorder 3G supports input only). Use this method to check device capabilities before attempting to use them for input or output.

initialize(device_index=0) -> bool Initialize the specified DeckLink device.

• device_index: Index of device to use (default: 0)
• Returns: True if successful

Note: Explicit initialization is optional. Methods like display_static_frame() and display_solid_color() will automatically initialize the first available device (device_index=0) if not already initialized. Explicit initialization is recommended when you need:

• To select a specific device (when multiple DeckLink devices are present)
• Separate error handling for device initialization vs. frame display
• Control over initialization timing (e.g., to avoid delays during first frame display)
• To verify device availability before preparing frame data

get_supported_display_modes() -> List[dict] Get list of supported display modes for the initialized device.

• Returns: List of dictionaries, each containing:
  • display_mode: DisplayMode enum value
  • name: Human-readable mode name (e.g., "1080p25", "2160p59.94")
  • width: Frame width in pixels
  • height: Frame height in pixels
  • framerate: Frame rate in frames per second
• Raises: RuntimeError if device not initialized

Example:

from blackmagic_io import BlackmagicOutput

with BlackmagicOutput() as output:
    output.initialize()

    modes = output.get_supported_display_modes()
    for mode in modes:
        print(f"{mode['name']}: {mode['width']}x{mode['height']} @ {mode['framerate']:.2f} fps")

is_pixel_format_supported(display_mode, pixel_format) -> bool Check if a pixel format is supported for a given display mode.

• display_mode: Display mode to check
• pixel_format: Pixel format to check
• Returns: True if the mode / format combination is supported

display_static_frame(frame_data, display_mode, pixel_format=PixelFormat.YUV10, matrix=None, hdr_metadata=None, input_narrow_range=False, output_narrow_range=True) -> bool Display a static frame continuously.

• frame_data: NumPy array with image data. Accepted shape and dtype depend on pixel_format:
  • BGRA (fast preview): shape (height, width, 3) R'G'B' or (height, width, 4) BGRA; dtype uint8 only. For float or higher bit depth, use one of the formats below — BGRA is converted to 8-bit Y'CbCr (2vuy) over SDI on tested hardware, so passing higher-precision data here causes double quantisation.
  • YUV8: shape (height, width, 3) R'G'B'; dtype uint8, uint16, float32, or float64.
  • YUV10 / RGB10 / RGB12: shape (height, width, 3) R'G'B'; dtype uint16, float32, or float64.
• display_mode: Video resolution and frame rate
• pixel_format: Pixel format (default: YUV10, automatically uses BGRA for uint8 data)
• matrix: Optional R'G'B' to Y'CbCr conversion matrix (Matrix.Rec601, Matrix.Rec709 or Matrix.Rec2020). Only used with YUV10 format. If not specified, auto-detects based on resolution: SD modes (NTSC, PAL) use Rec.601, HD and higher use Rec.709
• hdr_metadata: Optional HDR metadata dict with keys:
  • 'eotf': Eotf enum (SDR, PQ, or HLG)
  • 'static_metadata': Optional HdrStaticMetadata object with explicit display primaries, white point, mastering luminance, and content light level fields
• input_narrow_range: Whether to interpret integer frame_data as narrow range (float is always interpreted as full range). Default: False
• output_narrow_range: Whether to output a narrow range signal. Default: True
• Returns: True if successful

display_solid_color(color, display_mode, pixel_format=PixelFormat.YUV10, matrix=None, hdr_metadata=None, input_narrow_range=False, output_narrow_range=True, patch=None, background_color=None) -> bool Display a solid color continuously.

• color: R'G'B' tuple (r, g, b) with values:
  • Integer values (0-1023): Interpreted as 10-bit values
  • Float values (0.0-1.0): Interpreted as normalized full range values
• display_mode: Video resolution and frame rate
• pixel_format: Pixel format (default: YUV10)
• matrix: R'G'B' to Y'CbCr conversion matrix (Rec601, Rec709 or Rec2020). Only applies when pixel_format is YUV10. If not specified, auto-detects based on resolution: SD modes (NTSC, PAL) use Rec.601, HD and higher use Rec.709
• hdr_metadata: Optional HDR metadata dict with 'eotf' (and optional 'static_metadata') keys
• input_narrow_range: Whether to interpret integer color values as narrow range (float is always interpreted as full range). Default: False
• output_narrow_range: Whether to output a narrow range signal. Default: True
• patch: Optional tuple (center_x, center_y, width, height) with normalized coordinates (0.0-1.0):
  • center_x, center_y: Center position of the patch (0.5, 0.5 = center of screen)
  • width, height: Patch dimensions (1.0, 1.0 = full screen)
  • If None, displays full screen solid color. Default: None
• background_color: R'G'B' tuple for background when using patch parameter. Uses same format as color parameter (respecting input_narrow_range). If None, defaults to black. Default: None
• Returns: True if successful

update_frame(frame_data) -> bool Update currently displayed frame with new data.

• frame_data: New frame data as NumPy array
• Returns: True if successful

get_display_mode_info(display_mode) -> dict Get information about a display mode.

• Returns: Dictionary with 'width', 'height', 'framerate'

get_current_output_info() -> dict Get information about the current output configuration.

• Returns: Dictionary with 'display_mode_name', 'pixel_format_name', 'width', 'height', 'framerate', 'rgb444_mode_enabled'

stop() -> bool Stop video output.

• Returns: True if successful

Stops displaying frames but keeps the device initialized and ready for immediate reuse. After calling stop(), you can call display_static_frame() or display_solid_color() again without needing to re-initialize.

cleanup() Cleanup resources and stop output.

Stops video output (if running) and releases all device resources. After cleanup(), the device must be re-initialized with initialize() before it can be used again. This method automatically calls stop() internally, so there is no need to call stop() first.

Context Manager Support:

with BlackmagicOutput() as output:
    output.initialize()
    # ... use output ...
# Automatic cleanup() called on exit

The context manager automatically calls cleanup() when exiting, so explicit cleanup is not needed when using the with statement.

Utility Functions

create_test_pattern(width, height, pattern='gradient', grad_start=0.0, grad_end=1.0) -> np.ndarray Create test patterns for display testing and calibration.

• width: Frame width in pixels
• height: Frame height in pixels
• pattern: Pattern type - 'gradient', 'bars', or 'checkerboard'
• grad_start: Float starting value for gradient pattern (default: 0.0, use <0.0 for sub-black)
• grad_end: Float ending value for gradient pattern (default: 1.0, use >1.0 for super-white)
• Returns: R'G'B' array (H×W×3), dtype float32

High-Level API: BlackmagicInput Class

Convenient Python wrapper for video capture operations.

Methods

get_available_devices() -> List[str] Get list of available DeckLink device names.

get_device_capabilities(device_index=0) -> dict Get device capabilities (name and supported input/output).

• device_index: Index of device to query (default: 0)
• Returns: Dictionary with:
  • 'name': Device name
  • 'supports_input': True if device can capture video
  • 'supports_output': True if device can output video

get_available_input_connections(device_index=0) -> List[InputConnection] Get available input connections for a DeckLink device.

• device_index: Index of device to query (default: 0)
• Returns: List of InputConnection enum values (e.g., [InputConnection.SDI, InputConnection.HDMI])

Use this to check which physical inputs are available on a device before selecting one with initialize().

initialize(device_index=0, input_connection=None, pixel_format=None) -> bool Initialize the specified DeckLink device for input and start capture.

• device_index: Index of device to use (default: 0)
• input_connection: Optional InputConnection enum to select specific input (e.g., InputConnection.SDI, InputConnection.HDMI). If None, uses the device's current/default input.
• pixel_format: Optional PixelFormat to request from hardware. Use PixelFormat.BGRA for fast real-time preview (default: YUV10 for quality capture)
• Returns: True if successful

Immediately activates capture mode, which will:

• Start accepting input signal
• Activate front panel display (if present)
• Enable format detection

Performance Note: Requesting PixelFormat.BGRA enables real-time ~25fps preview by having the hardware deliver 8-bit BGRA frames directly, avoiding expensive colorspace conversions. This is ideal for monitoring and preview workflows. For quality capture workflows, use the default YUV10 format (or explicitly specify it) to capture full 10-bit precision, then use capture_frame_as_rgb() or capture_frame_with_metadata() for processing.

capture_frame_as_uint8(timeout_ms=5000, input_narrow_range=True, output_narrow_range=False) -> Optional[np.ndarray] Capture a single frame and convert to R'G'B' uint8 (faster than float conversion).

• timeout_ms: Timeout in milliseconds (default: 5000)
• input_narrow_range: Whether input uses narrow range encoding (default: True)
• output_narrow_range: If False (default), output uint8 is full range (0-255, "ready to display"). If True, output is narrow-range R'G'B' (16-235 per channel) — useful when feeding the result to further video processing that expects narrow-range conventions.
• Returns: R'G'B' uint8 array (H×W×3), or None if timeout/no signal
• Automatically converts from any DeckLink pixel format to R'G'B'
• Faster than capture_frame_as_rgb() due to uint8 output, ideal for preview workflows
• When the capture was initialised with pixel_format=PixelFormat.BGRA and the SDK delivers 10-bit R'G'B' (the typical case for 8-bit R'G'B' sources on HDMI), the library automatically right-shifts each channel by 2 to recover the exact 8-bit values before applying any range conversion.

capture_frame_as_uint8_with_metadata(timeout_ms=5000, input_narrow_range=True, output_narrow_range=False) -> Optional[dict] Capture a frame as R'G'B' uint8 with format metadata (fast preview with metadata access).

• timeout_ms: Timeout in milliseconds (default: 5000)
• input_narrow_range: Whether input uses narrow range encoding (default: True)
• output_narrow_range: If False (default), output uint8 is full range (0-255). If True, output is narrow-range R'G'B' (16-235 per channel).
• Returns: Dictionary with frame data and metadata, or None if timeout/no signal

Dictionary keys:

• 'rgb': R'G'B' uint8 array (H×W×3)
• 'width': Frame width in pixels
• 'height': Frame height in pixels
• 'format': Pixel format name (e.g., "YUV10", "RGB10")
• 'mode': Display mode name (e.g., "HD1080p25")
• 'colorspace': Color matrix name (e.g., "Rec709", "Rec2020")
• 'eotf': Transfer function name (e.g., "SDR", "PQ", "HLG")
• 'input_narrow_range': Boolean indicating the input range used for conversion
• 'output_narrow_range': Boolean indicating the output range applied
• 'hdr_metadata': Dictionary with HDR metadata (only present if HDR metadata is in the signal)
  • Same structure as capture_frame_with_metadata() below

This function combines the performance of capture_frame_as_uint8() with metadata access, making it ideal for real-time preview applications that need to detect signal changes (resolution, colorspace, EOTF) without the overhead of float conversion.

capture_frame_as_uint16(timeout_ms=5000, input_narrow_range=True, output_narrow_range=False) -> Optional[np.ndarray] Capture a single frame and convert to R'G'B' uint16 (preserves native bit depth of the source).

• timeout_ms: Timeout in milliseconds (default: 5000)
• input_narrow_range: Whether input uses narrow range encoding (default: True)
• output_narrow_range: If False (default), output uint16 is full range (0-65535 scaled). If True, output is narrow-range R'G'B' (10-bit narrow codes LSB-padded to 16-bit: 4096-60160).
• Returns: R'G'B' uint16 array (H×W×3), or None if timeout/no signal
• Higher-precision counterpart to capture_frame_as_uint8(). 10-bit (RGB10 / YUV10) and 12-bit (RGB12) sources keep their native precision in the uint16 result. 8-bit sources (BGRA, or RGB10-delivered-as-BGRA) are LSB-padded via << 8 — 0xff maps to 0xff00, so the underlying precision is still 8-bit even though the dtype is uint16.

capture_frame_as_uint16_with_metadata(timeout_ms=5000, input_narrow_range=True, output_narrow_range=False) -> Optional[dict] Capture a frame as R'G'B' uint16 with format metadata. Higher-precision counterpart to capture_frame_as_uint8_with_metadata(); see that method for the per-key dictionary structure (only the 'rgb' value's dtype changes from uint8 to uint16). See capture_frame_as_uint16() above for notes on bit-depth handling per source format.

capture_frame_as_rgb(timeout_ms=5000, input_narrow_range=True) -> Optional[np.ndarray] Capture a single frame and convert to R'G'B'.

• timeout_ms: Timeout in milliseconds (default: 5000)
• input_narrow_range: Whether input uses narrow range encoding (default: True)
• Returns: R'G'B' float32 array (H×W×3) mapped to 0.0-1.0 (legal black to legal white per input_narrow_range), or None if timeout/no signal
• Automatically converts from any DeckLink pixel format to R'G'B'
• When the capture was initialised with pixel_format=PixelFormat.BGRA and the SDK delivers 10-bit R'G'B' (the typical case for 8-bit R'G'B' sources on HDMI), the library automatically right-shifts each channel by 2 to recover the exact 8-bit values before float conversion. This avoids the small precision error that comes from decoding LSB-padded 8-bit content as if it were native 10-bit.

capture_frame_with_metadata(timeout_ms=5000, input_narrow_range=True) -> Optional[dict] Capture a frame with format metadata.

• timeout_ms: Timeout in milliseconds (default: 5000)
• input_narrow_range: Whether input uses narrow range encoding (default: True)
• Returns: Dictionary with frame data and metadata, or None if timeout/no signal
• BGRA-requested handling: same automatic right-shift as capture_frame_as_rgb() above.

Dictionary keys:

• 'rgb': R'G'B' float32 array (H×W×3), values 0.0-1.0
• 'width': Frame width in pixels
• 'height': Frame height in pixels
• 'format': Pixel format name (e.g., "YUV10", "RGB10")
• 'mode': Display mode name (e.g., "1080p25")
• 'colorspace': Color matrix name (e.g., "Rec709", "Rec2020")
• 'eotf': Transfer function name (e.g., "SDR", "PQ", "HLG")
• 'input_narrow_range': Boolean indicating if input was narrow range
• 'hdr_metadata': Dictionary with HDR metadata (only present if HDR metadata is in the signal)
  • 'display_primaries': Dictionary with display primaries (if present)
    • 'red': Dictionary with 'x' and 'y' chromaticity coordinates
    • 'green': Dictionary with 'x' and 'y' chromaticity coordinates
    • 'blue': Dictionary with 'x' and 'y' chromaticity coordinates
  • 'white_point': Dictionary with 'x' and 'y' chromaticity coordinates (if present)
  • 'mastering_luminance': Dictionary with 'max' and 'min' luminance in cd/m² (if present)
  • 'content_light': Dictionary with content light levels (if present)
    • 'max_cll': Maximum content light level in cd/m² (MaxCLL) (if present)
    • 'max_fall': Maximum frame average light level in cd/m² (MaxFALL) (if present)

get_detected_format() -> Optional[dict] Get information about the detected input signal.

• Returns: Dictionary with format information, or None if no signal

Dictionary keys:

• 'mode': Display mode name
• 'width': Frame width in pixels
• 'height': Frame height in pixels
• 'framerate': Frame rate in fps

cleanup() Cleanup resources and stop capture.

Context Manager Support:

with BlackmagicInput() as input_device:
    input_device.initialize()
    # ... use input ...
# Automatic cleanup() called on exit

The context manager automatically calls cleanup() when exiting.

Low-Level API: DeckLinkOutput Class

Direct C++ API for more fine-grained control.

Methods

get_device_list() -> List[str] Get list of available DeckLink devices.

initialize(device_index=0) -> bool Initialize the specified DeckLink device.

get_supported_display_modes() -> List[DisplayModeInfo] Get list of supported display modes for the initialized device.

• Returns: List of DisplayModeInfo objects with display_mode, name, width, height, framerate

is_pixel_format_supported(display_mode, pixel_format) -> bool Check if a pixel format is supported for a given display mode.

get_video_settings(display_mode) -> VideoSettings Get video settings object for a display mode.

set_hdr_metadata(colorimetry: Gamut, eotf: Eotf) Set HDR metadata with default values. Must be called before setup_output().

set_hdr_static_metadata(colorimetry: Gamut, eotf: Eotf, static_metadata: HdrStaticMetadata) Set HDR Static Metadata (per SMPTE ST 2086 / CEA-861.3 Type 1) with explicit display primaries, white point, mastering display luminance, and content light level fields. Must be called before setup_output().

clear_hdr_metadata() Clear HDR metadata and reset to SDR. Call before setup_output() if you want to ensure no HDR metadata is present.

HDMI vs SDI when changing HDR metadata mid-stream: SDI carries metadata per-frame in the VPID, so SDI consumers see updated metadata on the very next frame after a set_hdr_metadata*() call. On HDMI, the HDR Static Metadata InfoFrame is sticky — it does not update until new video data is sent — so after changing metadata mid-stream you must call display_frame() again for HDMI consumers to see the new values. The frame contents do not need to change; pushing the same frame is enough to refresh the InfoFrame.

setup_output(settings: VideoSettings) -> bool Setup output with detailed settings.

set_frame_data(data: np.ndarray) -> bool Set frame data from NumPy array (must be in correct format).

display_frame() -> bool Display the current frame synchronously. Call this after set_frame_data() to update the display.

get_current_output_info() -> OutputInfo Get information about the current output configuration.

• Returns: OutputInfo struct with display_mode_name, pixel_format_name, width, height, framerate, rgb444_mode_enabled

stop_output() -> bool Stop video output.

• Returns: True if successful

Stops displaying frames but keeps the device initialized and ready for immediate reuse. After calling stop_output(), you can call setup_output() and display_frame() again without needing to re-initialize.

cleanup() Cleanup all resources.

Stops video output (if running) and releases all device resources. After cleanup(), the device must be re-initialized with initialize() before it can be used again. This method automatically calls stop_output() internally, so there is no need to call stop_output() first.

Data Structures

VideoSettings

class VideoSettings:
    mode: DisplayMode      # Video mode (resolution / framerate)
    format: PixelFormat    # Pixel format
    width: int             # Frame width in pixels
    height: int            # Frame height in pixels
    framerate: float       # Frame rate (e.g., 25.0, 29.97, 60.0)
    colorimetry: Gamut     # Y'CbCr matrix (Rec601 / Rec709 / Rec2020)
    eotf: Eotf             # Transfer function (SDR / PQ / HLG)

Note: What the Blackmagic SDK refers to as the "color space" (BMDColorspace) is in fact the matrix used for R'G'B' to Y'CbCr conversion, not the gamut of the image data. For example, ARRI Wide Gamut data would typically be converted using a Rec.709 matrix.

HdrStaticMetadata

The fields described in SMPTE ST 2086 (mastering display) and CEA-861.3 (HDR Static Metadata Type 1 InfoFrame: MaxCLL, MaxFALL).

class HdrStaticMetadata:
    # Display primaries (xy chromaticity coordinates)
    display_primaries_red_x: float
    display_primaries_red_y: float
    display_primaries_green_x: float
    display_primaries_green_y: float
    display_primaries_blue_x: float
    display_primaries_blue_y: float
    white_point_x: float
    white_point_y: float

    # Luminance values (nits)
    max_display_mastering_luminance: float
    min_display_mastering_luminance: float
    max_content_light_level: float
    max_frame_average_light_level: float

OutputInfo

class OutputInfo:
    display_mode: DisplayMode         # Current display mode
    pixel_format: PixelFormat         # Current pixel format
    width: int                        # Frame width in pixels
    height: int                       # Frame height in pixels
    framerate: float                  # Frame rate (e.g., 25.0, 29.97, 60.0)
    rgb444_mode_enabled: bool         # Whether R'G'B' 4:4:4 mode is enabled
    display_mode_name: str            # Human-readable display mode name
    pixel_format_name: str            # Human-readable pixel format name

DisplayModeInfo

class DisplayModeInfo:
    display_mode: DisplayMode         # Display mode enum value
    name: str                         # Human-readable mode name
    width: int                        # Frame width in pixels
    height: int                       # Frame height in pixels
    framerate: float                  # Frame rate (e.g., 25.0, 29.97, 60.0)

DeviceCapabilities

class DeviceCapabilities:
    name: str                         # Device name
    supports_input: bool              # True if device can capture video
    supports_output: bool             # True if device can output video

Returned by get_device_capabilities() to query what a device supports before initializing it.

CapturedFrame

class CapturedFrame:
    # Frame data
    data: List[uint8]                 # Raw frame data
    width: int                        # Frame width in pixels
    height: int                       # Frame height in pixels
    format: PixelFormat               # Pixel format
    mode: DisplayMode                 # Display mode
    valid: bool                       # Whether frame is valid

    # Format metadata
    colorspace: Gamut                 # Color matrix (Rec601/Rec709/Rec2020)
    eotf: Eotf                        # Transfer function (SDR/PQ/HLG)
    has_metadata: bool                # Whether metadata is present

    # Timecode (if present)
    has_timecode: bool                # Whether timecode is present
    timecode_hours: int               # Timecode hours (0-23)
    timecode_minutes: int             # Timecode minutes (0-59)
    timecode_seconds: int             # Timecode seconds (0-59)
    timecode_frames: int              # Timecode frames (frame number within second)
    timecode_is_drop_frame: bool      # True for drop frame timecode

    # HDR metadata (if present)
    display_primaries_red_x: float
    display_primaries_red_y: float
    display_primaries_green_x: float
    display_primaries_green_y: float
    display_primaries_blue_x: float
    display_primaries_blue_y: float
    has_display_primaries: bool

    white_point_x: float
    white_point_y: float
    has_white_point: bool

    max_display_mastering_luminance: float
    min_display_mastering_luminance: float
    has_mastering_luminance: bool

    max_content_light_level: float
    has_max_cll: bool

    max_frame_average_light_level: float
    has_max_fall: bool

Used by the low-level DeckLinkInput.capture_frame() method. Contains raw frame data plus all detected metadata including timecode and HDR information.

Low-Level API: DeckLinkInput Class

Direct C++ API for more fine-grained control over video capture.

Methods

get_device_list() -> List[str] Get list of available DeckLink devices.

get_available_input_connections(device_index=0) -> List[InputConnection] Get available input connections for a DeckLink device.

• device_index: Index of device to query (default: 0)
• Returns: List of InputConnection enum values

initialize(device_index=0, input_connection=None) -> bool Initialize the specified DeckLink device for input.

• device_index: Index of device to use (default: 0)
• input_connection: Optional InputConnection enum to select specific input. If None, uses device's current/default input.
• Returns: True if successful

start_capture(format=PixelFormat.Format10BitYUV) -> bool Start capturing with specified or auto-detected format.

• format: Optional PixelFormat to request from hardware (default: Format10BitYUV)
• Returns: True if successful

Use PixelFormat.Format8BitBGRA for fast real-time preview workflows where 8-bit precision is acceptable. This avoids expensive colorspace conversions and enables ~25fps capture rates.

capture_frame(frame, timeout_ms=5000) -> bool Capture a single frame.

• frame: CapturedFrame object to populate
• timeout_ms: Timeout in milliseconds (default: 5000)
• Returns: True if successful

stop_capture() -> bool Stop video capture.

• Returns: True if successful

get_detected_format() -> VideoSettings Get the detected video format.

• Returns: VideoSettings object with format information

get_detected_pixel_format() -> PixelFormat Get the detected pixel format.

• Returns: PixelFormat enum value

get_video_settings(mode) -> VideoSettings Get video settings for a display mode.

get_supported_display_modes() -> List[DisplayModeInfo] Get list of supported display modes for the initialized device.

set_hdmi_input_dynamic_ranges(dynamic_range_mask: int) -> bool Set the BMDDynamicRange bitmask advertised in the HDMI input EDID. Sources read this to decide which transfer functions they may transmit. Pass any combination of BMDDynamicRange bits as a single int — values are passed through to the SDK so newer SDKs adding additional bits work without library changes.

• dynamic_range_mask: Bitwise OR of bmdDynamicRangeSDR (0), bmdDynamicRangeHDRStaticPQ (1 << 29), and/or bmdDynamicRangeHDRStaticHLG (1 << 30)
• Returns: True if the mask was stored or applied successfully

The library defaults to advertising SDR | HDR Static PQ | HDR Static HLG. The SDK default omits HLG, which causes many HDMI sources to strip HDR Static Metadata when transmitting HLG; the library's default fixes that. May be called before or after initialize(). Has no effect on non-HDMI connections or on hardware that does not expose IDeckLinkHDMIInputEDID (older devices) — these cases soft-fail and capture proceeds normally. The library releases its EDID interface in cleanup(), which restores the default EDID per the SDK.

cleanup() Cleanup and release resources.

Utility Functions

rgb_to_bgra(rgb_array, width, height) -> np.ndarray Convert RGB to BGRA format.

• rgb_array: NumPy array (H×W×3), dtype uint8
• 8-bit data is always treated as full range, but 8-bit Y'CbCr output will always be narrow range
• Returns: BGRA array (H×W×4), dtype uint8

rgb_uint8_to_yuv8(rgb_array, width, height, matrix=Matrix.Rec709, input_narrow_range=False, output_narrow_range=True) -> np.ndarray Convert R'G'B' uint8 to 8-bit Y'CbCr 2vuy format.

• rgb_array: NumPy array (H×W×3), dtype uint8 (0-255 range)
• matrix: R'G'B' to Y'CbCr conversion matrix (Matrix.Rec601, Matrix.Rec709 or Matrix.Rec2020). Default: Matrix.Rec709
• input_narrow_range: Whether to interpret the rgb_array as narrow range (16-235). Default: False
• output_narrow_range: Whether to encode the Y'CbCr as narrow range (Y: 16-235, CbCr: 16-240; clamped to [0, 255], so super-blacks/super-whites are preserved). Default: True
• Returns: Packed 2vuy array

rgb_uint16_to_yuv8(rgb_array, width, height, matrix=Matrix.Rec709, input_narrow_range=False, output_narrow_range=True) -> np.ndarray Convert R'G'B' uint16 to 8-bit Y'CbCr 2vuy format.

• rgb_array: NumPy array (H×W×3), dtype uint16 (0-65535 range)
• matrix: R'G'B' to Y'CbCr conversion matrix (Matrix.Rec601, Matrix.Rec709 or Matrix.Rec2020). Default: Matrix.Rec709
• input_narrow_range: Whether to interpret the rgb_array as narrow range. Default: False
• output_narrow_range: Whether to encode the Y'CbCr as narrow range (Y: 16-235, CbCr: 16-240; clamped to [0, 255], so super-blacks/super-whites are preserved). Default: True
• Returns: Packed 2vuy array

rgb_float_to_yuv8(rgb_array, width, height, matrix=Matrix.Rec709, output_narrow_range=True) -> np.ndarray Convert R'G'B' float to 8-bit Y'CbCr 2vuy format.

• rgb_array: NumPy array (H×W×3), dtype float32 (0.0-1.0 range)
• matrix: R'G'B' to Y'CbCr conversion matrix (Matrix.Rec601, Matrix.Rec709 or Matrix.Rec2020). Default: Matrix.Rec709
• output_narrow_range: Whether to encode the Y'CbCr as narrow range (Y: 16-235, CbCr: 16-240; clamped to [0, 255], so super-blacks/super-whites are preserved). Default: True
• Returns: Packed 2vuy array

rgb_uint16_to_yuv10(rgb_array, width, height, matrix=Matrix.Rec709, input_narrow_range=False, output_narrow_range=True) -> np.ndarray Convert R'G'B' uint16 to 10-bit Y'CbCr v210 format.

• rgb_array: NumPy array (H×W×3), dtype uint16 (0-65535 range)
• matrix: R'G'B' to Y'CbCr conversion matrix (Matrix.Rec601, Matrix.Rec709 or Matrix.Rec2020). Default: Matrix.Rec709
• input_narrow_range: Whether to interpret the rgb_array as narrow range. Default: False
• output_narrow_range: Whether to encode the Y'CbCr as narrow range. Default: True
• Returns: Packed v210 array

rgb_float_to_yuv10(rgb_array, width, height, matrix=Matrix.Rec709, output_narrow_range=True) -> np.ndarray Convert R'G'B' float to 10-bit Y'CbCr v210 format.

• rgb_array: NumPy array (H×W×3), dtype float32 (0.0-1.0 range)
• matrix: R'G'B' to Y'CbCr conversion matrix (Matrix.Rec601, Matrix.Rec709 or Matrix.Rec2020). Default: Matrix.Rec709
• output_narrow_range: Whether to encode the Y'CbCr as narrow range. Default: True
• Returns: Packed v210 array

rgb_uint16_to_rgb10(rgb_array, width, height, input_narrow_range=True, output_narrow_range=True) -> np.ndarray Convert R'G'B' uint16 to 10-bit R'G'B' (bmdFormat10BitRGBXLE) format.

• rgb_array: NumPy array (H×W×3), dtype uint16 (0-65535 range)
• input_narrow_range: Whether to interpret the rgb_array as narrow range. Default: True
• output_narrow_range: Whether to output narrow range. Default: True
• Returns: Packed 10-bit R'G'B' array

rgb_float_to_rgb10(rgb_array, width, height, output_narrow_range=True) -> np.ndarray Convert R'G'B' float to 10-bit R'G'B' (bmdFormat10BitRGBXLE) format.

• rgb_array: NumPy array (H×W×3), dtype float32 (0.0-1.0 range)
• output_narrow_range: Whether to output narrow range. Default: True
• Returns: Packed 10-bit R'G'B' array

rgb_uint16_to_rgb12(rgb_array, width, height, input_narrow_range=False, output_narrow_range=False) -> np.ndarray Convert R'G'B' uint16 to 12-bit R'G'B' (bmdFormat12BitRGBLE) format.

• rgb_array: NumPy array (H×W×3), dtype uint16 (0-65535 range)
• input_narrow_range: Whether to interpret the rgb_array as narrow range. Default: False
• output_narrow_range: Whether to output narrow range. Default: False
• Returns: Packed 12-bit R'G'B' array

rgb_float_to_rgb12(rgb_array, width, height, output_narrow_range=False) -> np.ndarray Convert R'G'B' float to 12-bit R'G'B' (bmdFormat12BitRGBLE) format.

• rgb_array: NumPy array (H×W×3), dtype float32 (0.0-1.0 range)
• output_narrow_range: Whether to output narrow range. Default: False
• Returns: Packed 12-bit R'G'B' array

yuv10_to_rgb_uint16(yuv_array, width, height, matrix=Matrix.Rec709, input_narrow_range=True, output_narrow_range=False, row_bytes=None) -> np.ndarray Convert 10-bit Y'CbCr v210 format to R'G'B' uint16.

• yuv_array: NumPy array containing packed v210 data
• width: Frame width in pixels
• height: Frame height in pixels
• matrix: Y'CbCr to R'G'B' conversion matrix (Matrix.Rec601, Matrix.Rec709 or Matrix.Rec2020). Default: Matrix.Rec709
• input_narrow_range: Whether to interpret the Y'CbCr as narrow range. Default: True
• output_narrow_range: Whether to encode the uint16 R'G'B' output as narrow range (4096-60160 @ 16-bit). Default: False
• row_bytes: Bytes per row in the source buffer. Pass captured_frame.row_bytes for captured frames whose stride may include padding. If None, defaults to ((width + 47) // 48) * 128 (the v210-native stride).
• Returns: NumPy array (H×W×3), dtype uint16 (0-65535 range)

yuv10_to_rgb_float(yuv_array, width, height, matrix=Matrix.Rec709, input_narrow_range=True, row_bytes=None) -> np.ndarray Convert 10-bit Y'CbCr v210 format to R'G'B' float.

• yuv_array: NumPy array containing packed v210 data
• width: Frame width in pixels
• height: Frame height in pixels
• matrix: Y'CbCr to R'G'B' conversion matrix (Matrix.Rec601, Matrix.Rec709 or Matrix.Rec2020). Default: Matrix.Rec709
• input_narrow_range: Whether to interpret the Y'CbCr as narrow range. Default: True
• row_bytes: Bytes per row in the source buffer. Pass captured_frame.row_bytes for captured frames whose stride may include padding. If None, defaults to ((width + 47) // 48) * 128 (the v210-native stride).
• Returns: NumPy array (H×W×3), dtype float32 (0.0-1.0 range)

yuv8_to_rgb_uint16(yuv_array, width, height, matrix=Matrix.Rec709, input_narrow_range=True, output_narrow_range=False, row_bytes=None) -> np.ndarray Convert 8-bit Y'CbCr 4:2:2 (2vuy) format to R'G'B' uint16.

• yuv_array: NumPy array containing packed 2vuy data
• width: Frame width in pixels
• height: Frame height in pixels
• matrix: Y'CbCr to R'G'B' conversion matrix (Matrix.Rec601, Matrix.Rec709 or Matrix.Rec2020). Default: Matrix.Rec709
• input_narrow_range: Whether to interpret the Y'CbCr as narrow range (Y: 16-235, CbCr: 16-240). Default: True
• output_narrow_range: Whether to encode the uint16 R'G'B' output as narrow range (4096-60160 @ 16-bit). Default: False
• row_bytes: Bytes per row in the source buffer. Pass captured_frame.row_bytes for captured frames whose stride may include padding. If None, defaults to width * 2.
• Returns: NumPy array (H×W×3), dtype uint16 (0-65535 range)

yuv8_to_rgb_float(yuv_array, width, height, matrix=Matrix.Rec709, input_narrow_range=True, row_bytes=None) -> np.ndarray Convert 8-bit Y'CbCr 4:2:2 (2vuy) format to R'G'B' float.

• yuv_array: NumPy array containing packed 2vuy data
• width: Frame width in pixels
• height: Frame height in pixels
• matrix: Y'CbCr to R'G'B' conversion matrix (Matrix.Rec601, Matrix.Rec709 or Matrix.Rec2020). Default: Matrix.Rec709
• input_narrow_range: Whether to interpret the Y'CbCr as narrow range (Y: 16-235, CbCr: 16-240). Default: True
• row_bytes: Bytes per row in the source buffer. Pass captured_frame.row_bytes for captured frames whose stride may include padding. If None, defaults to width * 2.
• Returns: NumPy array (H×W×3), dtype float32 (0.0-1.0 range)

rgb10_to_uint16(rgb_array, width, height, input_narrow_range=True, output_narrow_range=False, row_bytes=None) -> np.ndarray Convert 10-bit R'G'B' (bmdFormat10BitRGBXLE) format to R'G'B' uint16.

• rgb_array: NumPy array containing packed 10-bit R'G'B' data
• width: Frame width in pixels
• height: Frame height in pixels
• input_narrow_range: Whether to interpret the 10-bit R'G'B' as narrow range. Default: True
• output_narrow_range: Whether to output narrow range. Default: False
• row_bytes: Bytes per row in the source buffer. Pass captured_frame.row_bytes for captured frames whose stride may include padding. If None, defaults to width * 4.
• Returns: NumPy array (H×W×3), dtype uint16 (0-65535 range)

rgb10_to_float(rgb_array, width, height, input_narrow_range=True, row_bytes=None) -> np.ndarray Convert 10-bit R'G'B' (bmdFormat10BitRGBXLE) format to R'G'B' float.

• rgb_array: NumPy array containing packed 10-bit R'G'B' data
• width: Frame width in pixels
• height: Frame height in pixels
• input_narrow_range: Whether to interpret the 10-bit R'G'B' as narrow range. Default: True
• row_bytes: Bytes per row in the source buffer. Pass captured_frame.row_bytes for captured frames whose stride may include padding. If None, defaults to width * 4.
• Returns: NumPy array (H×W×3), dtype float32 (0.0-1.0 range)

rgb12_to_uint16(rgb_array, width, height, input_narrow_range=False, output_narrow_range=False, row_bytes=None) -> np.ndarray Convert 12-bit R'G'B' (bmdFormat12BitRGBLE) format to R'G'B' uint16.

• rgb_array: NumPy array containing packed 12-bit R'G'B' data
• width: Frame width in pixels
• height: Frame height in pixels
• input_narrow_range: Whether to interpret the 12-bit R'G'B' as narrow range. Default: False
• output_narrow_range: Whether to output narrow range. Default: False
• row_bytes: Bytes per row in the source buffer. Pass captured_frame.row_bytes for captured frames whose stride may include padding. If None, defaults to ((width + 7) // 8) * 36 (the R12L-native stride).
• Returns: NumPy array (H×W×3), dtype uint16 (0-65535 range)

rgb12_to_float(rgb_array, width, height, input_narrow_range=False, row_bytes=None) -> np.ndarray Convert 12-bit R'G'B' (bmdFormat12BitRGBLE) format to R'G'B' float.

• rgb_array: NumPy array containing packed 12-bit R'G'B' data
• width: Frame width in pixels
• height: Frame height in pixels
• input_narrow_range: Whether to interpret the 12-bit R'G'B' as narrow range. Default: False
• row_bytes: Bytes per row in the source buffer. Pass captured_frame.row_bytes for captured frames whose stride may include padding. If None, defaults to ((width + 7) // 8) * 36 (the R12L-native stride).
• Returns: NumPy array (H×W×3), dtype float32 (0.0-1.0 range)

unpack_v210(v210_array, width, height, row_bytes=None) -> dict Unpack 10-bit Y'CbCr v210 format to Y', Cb, Cr component arrays.

• v210_array: NumPy array containing packed v210 data
• width: Frame width in pixels
• height: Frame height in pixels
• row_bytes: Bytes per row in the source buffer. Pass captured_frame.row_bytes for captured frames whose stride may include padding. If None, defaults to ((width + 47) // 48) * 128 (the v210-native stride).
• Returns: Dictionary with 'y', 'cb', 'cr' keys, each containing an H×W NumPy array, dtype uint16 (0-1023 range, 10-bit values)

unpack_2vuy(yuv_array, width, height, row_bytes=None) -> dict Unpack 8-bit Y'CbCr 4:2:2 (2vuy) format to Y', Cb, Cr component arrays.

• yuv_array: NumPy array containing packed 2vuy data
• width: Frame width in pixels
• height: Frame height in pixels
• row_bytes: Bytes per row in the source buffer. Pass captured_frame.row_bytes for captured frames whose stride may include padding. If None, defaults to width * 2.
• Returns: Dictionary with 'y', 'cb', 'cr' keys, each containing an H×W NumPy array, dtype uint8

unpack_rgb10(rgb_array, width, height, row_bytes=None) -> dict Unpack 10-bit R'G'B' (bmdFormat10BitRGBXLE) format to R', G', B' component arrays.

• rgb_array: NumPy array containing packed 10-bit R'G'B' data
• width: Frame width in pixels
• height: Frame height in pixels
• row_bytes: Bytes per row in the source buffer. Pass captured_frame.row_bytes for captured frames whose stride may include padding. If None, defaults to width * 4.
• Returns: Dictionary with 'r', 'g', 'b' keys, each containing an H×W NumPy array, dtype uint16 (0-1023 range, 10-bit values)

unpack_rgb12(rgb_array, width, height, row_bytes=None) -> dict Unpack 12-bit R'G'B' (bmdFormat12BitRGBLE) format to R', G', B' component arrays.

• rgb_array: NumPy array containing packed 12-bit R'G'B' data
• width: Frame width in pixels
• height: Frame height in pixels
• row_bytes: Bytes per row in the source buffer. Pass captured_frame.row_bytes for captured frames whose stride may include padding. If None, defaults to ((width + 7) // 8) * 36 (the R12L-native stride).
• Returns: Dictionary with 'r', 'g', 'b' keys, each containing an H×W NumPy array, dtype uint16 (0-4095 range, 12-bit values)

Enums

DisplayMode

The library supports all display modes available on your DeckLink device. Display mode settings (resolution, framerate) are queried dynamically from the hardware. Common examples include:

• HD1080p25: 1920×1080 @ 25fps
• HD1080p30: 1920×1080 @ 30fps
• HD1080p50: 1920×1080 @ 50fps
• HD1080p60: 1920×1080 @ 60fps
• HD720p50: 1280×720 @ 50fps
• HD720p60: 1280×720 @ 60fps

Additional modes are available including SD (NTSC, PAL), 2K, 4K, 8K, and PC display modes. The complete list of DisplayMode values can be found in src/blackmagic_io/blackmagic_io.py.

Querying Available Display Modes:

You can query which display modes are supported by your specific DeckLink device using get_supported_display_modes():

from blackmagic_io import BlackmagicOutput

with BlackmagicOutput() as output:
    output.initialize()

    # Get all supported display modes
    modes = output.get_supported_display_modes()
    print(f"Device supports {len(modes)} display modes:\n")
    for mode in modes:
        print(f"{mode['name']}: {mode['width']}x{mode['height']} @ {mode['framerate']:.2f} fps")

To determine which pixel formats are supported for a specific display mode, use is_pixel_format_supported():

from blackmagic_io import BlackmagicOutput, DisplayMode, PixelFormat

with BlackmagicOutput() as output:
    output.initialize()

    # Test pixel format support for a specific mode
    print("Pixel formats supported for HD1080p25:")
    test_formats = [PixelFormat.YUV10, PixelFormat.RGB10, PixelFormat.RGB12]
    for fmt in test_formats:
        supported = output.is_pixel_format_supported(DisplayMode.HD1080p25, fmt)
        status = "✓" if supported else "✗"
        print(f"{status} {fmt.name}")

PixelFormat

• BGRA: 8-bit BGRA (automatically used for uint8 data)
  • Note: Over SDI, BGRA data is output as narrow range 8-bit Y'CbCr 4:2:2, not R'G'B'. The BGRA name refers to the input buffer format, not the SDI wire format.
• YUV8: 8-bit Y'CbCr 4:2:2 (2vuy) - direct 8-bit Y'CbCr output
  • uint8 input: Configurable interpretation via input_narrow_range parameter
  • uint16 input: Configurable interpretation via input_narrow_range parameter
  • float input: Always interpreted as full range (0.0-1.0)
  • Output range configurable via output_narrow_range parameter
  • Defaults: input_narrow_range=False, output_narrow_range=True
  • More efficient than BGRA for 8-bit Y'CbCr workflows (avoids hardware conversion)
• YUV10: 10-bit Y'CbCr 4:2:2 (v210) - default for uint16 / float data
  • Defaults to narrow range: Y: 64-940, UV: 64-960. Supports full range Y'CbCr (0-1023, as per Rec. ITU-T H.273) if output_narrow_range is False in the high level API
• RGB10: 10-bit R'G'B' (bmdFormat10BitRGBXLE) - native R'G'B' output without Y'CbCr conversion
  • uint16 input: Configurable interpretation via input_narrow_range parameter
  • float input: Always interpreted as full range (0.0-1.0)
  • Output range configurable via output_narrow_range parameter
  • Defaults: input_narrow_range=True, output_narrow_range=True
• RGB12: 12-bit R'G'B' (bmdFormat12BitRGBLE) - native R'G'B' output with 12-bit precision
  • uint16 input: Configurable interpretation via input_narrow_range parameter
  • float input: Always interpreted as full range (0.0-1.0)
  • Output range configurable via output_narrow_range parameter
  • Defaults: input_narrow_range=False, output_narrow_range=False

Range Signaling Limitations

Important: While this library supports both narrow and full range output encoding via the output_narrow_range parameter, the Blackmagic DeckLink SDK (v14.1) does not provide APIs to control the full range flag in the VPID, as per SMPTE ST 425-1 (byte 4, bit 7):

• YUV10: The library can encode full range Y'CbCr (0-1023) with output_narrow_range=False, but cannot set the full range flag in the VPID. Downstream devices may well assume narrow range.

• RGB10: The convention is that 10-bit R'G'B' is narrow range, as described in the Blackmagic SDK, so using output_narrow_range=False may cause downstream devices to misinterpret the signal.

• RGB12: The convention is that 12-bit R'G'B' is full range, as described in the Blackmagic SDK, so using output_narrow_range=True may cause downstream devices to misinterpret the signal.

The output_narrow_range parameter controls the actual encoded values in the output stream, not metadata signaling. Use it when you know the downstream device will correctly interpret the range, or when the receiving device allows manual range configuration.

Matrix (High-level API)

• Rec709: ITU-R BT.709 R'G'B' to Y'CbCr conversion matrix (standard HD)
• Rec2020: ITU-R BT.2020 R'G'B' to Y'CbCr conversion matrix (wide color gamut for HDR)

Gamut (Low-level API, same values as Matrix)

• Rec709: ITU-R BT.709 colorimetry (standard HD)
• Rec2020: ITU-R BT.2020 colorimetry (wide color gamut for HDR)

Eotf

• SDR: Standard Dynamic Range (BT.1886 transfer function)
• PQ: Perceptual Quantizer (SMPTE ST 2084, HDR10)
• HLG: Hybrid Log-Gamma (HDR broadcast standard)

InputConnection

Physical input connections available on DeckLink devices. Use get_available_input_connections() to query which inputs are available on a specific device.

• SDI: SDI input (Serial Digital Interface)
• HDMI: HDMI input
• OpticalSDI: Optical SDI input
• Component: Component video input (Y, Pb, Pr)
• Composite: Composite video input (CVBS)
• SVideo: S-Video input (Y/C)

Example:

from blackmagic_io import BlackmagicInput, InputConnection

with BlackmagicInput() as input_device:
    # Query available inputs
    inputs = input_device.get_available_input_connections(0)
    print(f"Available inputs: {[str(inp) for inp in inputs]}")

    # Initialize with HDMI input
    if InputConnection.HDMI in inputs:
        input_device.initialize(0, InputConnection.HDMI)
    else:
        # Use default input
        input_device.initialize(0)

    # Capture frame
    rgb = input_device.capture_frame_as_rgb()

Examples

Example 1: Color Bars Test Pattern

from blackmagic_io import BlackmagicOutput, DisplayMode, create_test_pattern

# Create color bars test pattern
frame = create_test_pattern(1920, 1080, 'bars')

with BlackmagicOutput() as output:
    output.initialize()  # Optional - auto-initializes on first display if omitted
    output.display_static_frame(frame, DisplayMode.HD1080p25)
    input("Press Enter to stop...")

Example 2: Dynamic Animation

import numpy as np
import time
from blackmagic_io import BlackmagicOutput, DisplayMode

with BlackmagicOutput() as output:
    # Start with black frame
    frame = np.zeros((1080, 1920, 3), dtype=np.float32)
    output.display_static_frame(frame, DisplayMode.HD1080p25)

    # Animate
    for i in range(100):
        # Create moving pattern
        frame.fill(0.0)
        offset = i * 10
        frame[:, offset:offset+100] = [1.0, 1.0, 1.0]  # White bar

        output.update_frame(frame)
        time.sleep(1 / 25)  # Limit update rate (actual rate will be lower due to processing overhead)

Example 3: Load Image from File

import imageio.v3 as iio
import numpy as np
from blackmagic_io import BlackmagicOutput, DisplayMode

# Load image (preserves bit depth for 16-bit TIFFs, etc.)
# Note: Use TIFF for reliable 16-bit support (PNGs may convert to 8-bit)
frame = iio.imread('your_image.tif')

# Resize if needed
if frame.shape[0] != 1080 or frame.shape[1] != 1920:
    from PIL import Image
    img = Image.fromarray(frame)
    img = img.resize((1920, 1080), Image.Resampling.LANCZOS)
    frame = np.array(img)

# Remove alpha channel if present
if frame.shape[2] == 4:
    frame = frame[:, :, :3]

# Display image (format auto-detected from dtype)
with BlackmagicOutput() as output:
    output.display_static_frame(frame, DisplayMode.HD1080p25)
    input("Press Enter to stop...")

Example 4: 10-bit Y'CbCr Output with Float Data

import numpy as np
from blackmagic_io import BlackmagicOutput, DisplayMode

# Create float R'G'B' image (0.0-1.0 range)
frame = np.zeros((1080, 1920, 3), dtype=np.float32)

# Example: gradient in float space
for y in range(1080):
    for x in range(1920):
        frame[y, x] = [
            x / 1920,           # Red gradient
            y / 1080,           # Green gradient
            0.5                 # Blue constant
        ]

# Output as 10-bit Y'CbCr (automatically selected for float data)
with BlackmagicOutput() as output:
    output.display_static_frame(frame, DisplayMode.HD1080p25)
    input("Press Enter to stop...")

Example 5: 10-bit Y'CbCr with uint16 Data

import numpy as np
from blackmagic_io import BlackmagicOutput, DisplayMode

# Create uint16 R'G'B' image (0-65535 range)
# Useful for 10-bit / 12-bit / 16-bit image processing pipelines
frame = np.zeros((1080, 1920, 3), dtype=np.uint16)

# Full range gradient
for x in range(1920):
    frame[:, x, 0] = int(x / 1920 * 65535)  # Red gradient

# Output as 10-bit Y'CbCr (automatically selected for uint16 data)
with BlackmagicOutput() as output:
    output.display_static_frame(frame, DisplayMode.HD1080p25)
    input("Press Enter to stop...")

Example 5a: 10-bit R'G'B' with uint16 Data

import numpy as np
from blackmagic_io import BlackmagicOutput, DisplayMode, PixelFormat

# Create uint16 R'G'B' image (0-65535 range)
frame = np.zeros((1080, 1920, 3), dtype=np.uint16)

# Full range gradient
for x in range(1920):
    frame[:, x, 0] = int(x / 1920 * 65535)  # Red gradient

# Output as 10-bit R'G'B' (bit-shifted from 16-bit to 10-bit)
with BlackmagicOutput() as output:
    output.display_static_frame(frame, DisplayMode.HD1080p25, PixelFormat.RGB10)
    input("Press Enter to stop...")

Example 5b: 10-bit R'G'B' with Float Data (Narrow Range)

import numpy as np
from blackmagic_io import BlackmagicOutput, DisplayMode, PixelFormat

# Create float R'G'B' image (0.0-1.0 range)
frame = np.zeros((1080, 1920, 3), dtype=np.float32)

# Gradient
for x in range(1920):
    frame[:, x, 0] = x / 1920  # Red gradient

# Output as 10-bit R'G'B' with narrow range (0.0-1.0 maps to 64-940)
with BlackmagicOutput() as output:
    output.display_static_frame(
        frame,
        DisplayMode.HD1080p25,
        PixelFormat.RGB10,
        output_narrow_range=True  # Default: narrow range
    )
    input("Press Enter to stop...")

Example 5c: 10-bit R'G'B' with Float Data (Full Range)

import numpy as np
from blackmagic_io import BlackmagicOutput, DisplayMode, PixelFormat

# Create float R'G'B' image (0.0-1.0 range)
frame = np.zeros((1080, 1920, 3), dtype=np.float32)

# Gradient
for x in range(1920):
    frame[:, x, 0] = x / 1920  # Red gradient

# Output as 10-bit R'G'B' with full range (0.0-1.0 maps to 0-1023)
with BlackmagicOutput() as output:
    output.display_static_frame(
        frame,
        DisplayMode.HD1080p25,
        PixelFormat.RGB10,
        output_narrow_range=False  # Full range
    )
    input("Press Enter to stop...")

Example 6: 8-bit BGRA Output

For simple applications or quick testing, 8-bit R'G'B' data can be used directly without conversion to float or uint16. Note that 8-bit data is always treated as full range R'G'B' input and output as narrow range 8-bit Y'CbCr 4:2:2 over SDI.

import numpy as np
from blackmagic_io import BlackmagicOutput, DisplayMode

# Create 8-bit R'G'B' image (0-255 range, full range)
frame = np.zeros((1080, 1920, 3), dtype=np.uint8)

# Simple gradient
for x in range(1920):
    frame[:, x, 0] = int(255 * x / 1920)  # Red gradient

# Output as 8-bit (BGRA format automatically selected for uint8 data)
with BlackmagicOutput() as output:
    output.display_static_frame(frame, DisplayMode.HD1080p25)
    input("Press Enter to stop...")

Example 7: HDR Output with Rec.2020 and HLG (Simplified API)

import numpy as np
from blackmagic_io import BlackmagicOutput, DisplayMode, Matrix, Eotf

# Create HDR content in normalised float (0.0-1.0 range)
frame = np.zeros((1080, 1920, 3), dtype=np.float32)

# Example: HDR gradient with extended range
for y in range(1080):
    for x in range(1920):
        frame[y, x] = [
            x / 1920,           # Red gradient
            y / 1080,           # Green gradient
            0.5                 # Blue constant
        ]

# Configure for HLG HDR output using the simplified API
with BlackmagicOutput() as output:
    # Single call with matrix and HDR metadata
    # YUV10 automatically selected for float data
    output.display_static_frame(
        frame,
        DisplayMode.HD1080p25,
        matrix=Matrix.Rec2020,           # Use Rec.2020 matrix
        hdr_metadata={'eotf': Eotf.HLG}  # HLG with default metadata
    )

    input("Press Enter to stop...")

Alternative: Low-level API for more control

import numpy as np
import decklink_io as dl

# Create HDR content
frame = np.zeros((1080, 1920, 3), dtype=np.float32)
# ... fill frame ...

# Configure for HLG HDR output using low-level API
output = dl.DeckLinkOutput()
output.initialize()

# Set HDR metadata BEFORE setup_output()
output.set_hdr_metadata(dl.Gamut.Rec2020, dl.Eotf.HLG)

# Setup output
settings = output.get_video_settings(dl.DisplayMode.HD1080p25)
settings.format = dl.PixelFormat.YUV10
output.setup_output(settings)

# Convert R'G'B' to Y'CbCr using Rec.2020 matrix
yuv_data = dl.rgb_float_to_yuv10(frame, 1920, 1080, dl.Matrix.Rec2020)
output.set_frame_data(yuv_data)

# Display the frame
output.display_frame()
input("Press Enter to stop...")
output.stop_output()
output.cleanup()

Example 8: HDR10 (PQ) Output (Simplified API)

import numpy as np
from blackmagic_io import BlackmagicOutput, DisplayMode, Matrix, Eotf

# Create HDR10 content with PQ transfer function applied
frame = np.zeros((1080, 1920, 3), dtype=np.float32)

# Fill with PQ-encoded HDR content
# A library such as Colour Science for Python (https://www.colour-science.org/) is needed for PQ encoding
# frame = colour.eotf(linear_rgb_data, 'ST 2084')

# Configure for HDR10 (PQ) output using the simplified API
with BlackmagicOutput() as output:
    # Single call with Rec.2020 matrix and PQ metadata
    # YUV10 automatically selected for float data
    output.display_static_frame(
        frame,
        DisplayMode.HD1080p25,
        matrix=Matrix.Rec2020,
        hdr_metadata={'eotf': Eotf.PQ}
    )

    input("Press Enter to stop...")

Alternative: Low-level API

import numpy as np
import decklink_io as dl

# Create HDR10 content
frame = np.zeros((1080, 1920, 3), dtype=np.float32)
# ... fill frame ...

# Configure for HDR10 (PQ) output
output = dl.DeckLinkOutput()
output.initialize()

# IMPORTANT: Set HDR metadata BEFORE setup_output()
# This embeds HDR metadata (including Rec.2020 primaries, EOTF, mastering display info) in frames
output.set_hdr_metadata(dl.Gamut.Rec2020, dl.Eotf.PQ)

# Setup output settings
settings = output.get_video_settings(dl.DisplayMode.HD1080p25)
settings.format = dl.PixelFormat.YUV10
output.setup_output(settings)

# Convert to Y'CbCr with Rec.2020 matrix
yuv_data = dl.rgb_float_to_yuv10(frame, 1920, 1080, dl.Gamut.Rec2020)
output.set_frame_data(yuv_data)

# Display the frame
output.display_frame()
input("Press Enter to stop...")
output.stop_output()
output.cleanup()

Example 9: Color Patches for Testing and Calibration

The display_solid_color() method supports displaying color patches smaller than full screen, useful for testing, calibration, and creating custom test patterns.

import time
from blackmagic_io import BlackmagicOutput, DisplayMode

with BlackmagicOutput() as output:
    # Full screen white (default behavior)
    output.display_solid_color((1.0, 1.0, 1.0), DisplayMode.HD1080p25)
    time.sleep(2)

    # Centered 50% white patch on black background
    output.display_solid_color(
        (1.0, 1.0, 1.0),
        DisplayMode.HD1080p25,
        patch=(0.5, 0.5, 0.5, 0.5)  # (center_x, center_y, width, height)
    )
    time.sleep(2)

    # Small centered white patch (10% size) on gray background
    output.display_solid_color(
        (1.0, 1.0, 1.0),
        DisplayMode.HD1080p25,
        patch=(0.5, 0.5, 0.1, 0.1),
        background_color=(0.5, 0.5, 0.5)
    )
    time.sleep(2)

    # Red patch in top-left quadrant on blue background
    output.display_solid_color(
        (1.0, 0.0, 0.0),
        DisplayMode.HD1080p25,
        patch=(0.25, 0.25, 0.3, 0.3),
        background_color=(0.0, 0.0, 1.0)
    )
    time.sleep(2)

    # Horizontal bar across center (full width, half height)
    # Using integer 10-bit values with narrow range
    output.display_solid_color(
        (940, 940, 64),
        DisplayMode.HD1080p25,
        patch=(0.5, 0.5, 1.0, 0.5),
        background_color=(400, 400, 400),
        input_narrow_range=True
    )
    time.sleep(2)

Patch coordinates:

• All values are normalized (0.0-1.0) for resolution independence
• center_x, center_y: Position of patch center (0.0 = left/top, 1.0 = right/bottom)
• width, height: Patch dimensions as fraction of screen (1.0 = full width/height)
• Example: (0.5, 0.5, 0.25, 0.25) = centered patch, 25% of screen size

Background color:

• Uses same format as foreground color (integers 0-1023 or floats 0.0-1.0)
• Defaults to black if not specified
• For integer colors with input_narrow_range=True, black defaults to 64 instead of 0

HDR Metadata

HDR metadata is embedded into each video frame using the DeckLink SDK's IDeckLinkVideoFrameMetadataExtensions interface. When you call set_hdr_metadata(), the library automatically wraps each output frame with the specified metadata.

Metadata Includes:

• Display Primaries: Automatically set to match the matrix parameter (unless explicitly specified via HdrStaticMetadata)
  • Matrix.Rec709 → Rec.709 primaries (x,y): R(0.64, 0.33), G(0.30, 0.60), B(0.15, 0.06)
  • Matrix.Rec2020 → Rec.2020 primaries (x,y): R(0.708, 0.292), G(0.170, 0.797), B(0.131, 0.046)
• White Point: D65 (0.3127, 0.3290) for all matrices (unless explicitly specified)
• EOTF: Electro-Optical Transfer Function (SDR / Rec.709, PQ / SMPTE ST 2084, or HLG)
• Mastering Display Info: Default values for max / min luminance
• Content Light Levels: Max content light level and max frame average

Default HDR Metadata Values (PQ only):

When using Matrix.Rec709:

Display Primaries: Rec.709 (ITU-R BT.709)
  Red:   (0.64, 0.33)
  Green: (0.30, 0.60)
  Blue:  (0.15, 0.06)
White Point: D65 (0.3127, 0.3290)

When using Matrix.Rec2020:

Display Primaries: Rec.2020 (ITU-R BT.2020)
  Red:   (0.708, 0.292)
  Green: (0.170, 0.797)
  Blue:  (0.131, 0.046)
White Point: D65 (0.3127, 0.3290)

Luminance values:

Max Mastering Luminance: 1000 nits
Min Mastering Luminance: 0.0001 nits
Max Content Light Level: 1000 nits
Max Frame Average Light Level: 50 nits

Customizing HDR Metadata Values:

High-level API:

from blackmagic_io import BlackmagicOutput, DisplayMode, PixelFormat, Matrix, Eotf
import decklink_io as dl

# Build HDR Static Metadata for the source's mastering display
static_metadata = dl.HdrStaticMetadata()
static_metadata.display_primaries_red_x = 0.708
static_metadata.display_primaries_red_y = 0.292
# ... set other values ...
static_metadata.max_display_mastering_luminance = 1000.0
static_metadata.min_display_mastering_luminance = 0.0001

# Use in simplified API
with BlackmagicOutput() as output:
    output.initialize()
    output.display_static_frame(
        frame,
        DisplayMode.HD1080p25,
        pixel_format=PixelFormat.YUV10,
        matrix=Matrix.Rec2020,
        hdr_metadata={'eotf': Eotf.PQ, 'static_metadata': static_metadata}
    )
    input("Press Enter to stop...")

Low-level API:

For precise control over HDR Static Metadata with the low-level API, use set_hdr_static_metadata():

import decklink_io as dl

# Build HDR Static Metadata for the mastering display
static_metadata = dl.HdrStaticMetadata()

# Display primaries (chromaticity coordinates)
static_metadata.display_primaries_red_x = 0.708
static_metadata.display_primaries_red_y = 0.292
static_metadata.display_primaries_green_x = 0.170
static_metadata.display_primaries_green_y = 0.797
static_metadata.display_primaries_blue_x = 0.131
static_metadata.display_primaries_blue_y = 0.046
static_metadata.white_point_x = 0.3127
static_metadata.white_point_y = 0.3290

# Mastering display luminance
static_metadata.max_display_mastering_luminance = 4000.0      # 4000 nits peak (e.g., for HDR10+ content)
static_metadata.min_display_mastering_luminance = 0.0005      # 0.0005 nits black level

# Content light levels
static_metadata.max_content_light_level = 2000.0     # 2000 nits max content (MaxCLL)
static_metadata.max_frame_average_light_level = 400.0 # 400 nits average (MaxFALL)

output = dl.DeckLinkOutput()
output.initialize()
output.set_hdr_static_metadata(dl.Gamut.Rec2020, dl.Eotf.PQ, static_metadata)

Available HDR Metadata Fields:

All 14 SMPTE ST 2086 / CEA-861.3 HDR static metadata fields are supported:

Display Primaries (xy chromaticity coordinates):

• display_primaries_red_x, display_primaries_red_y
• display_primaries_green_x, display_primaries_green_y
• display_primaries_blue_x, display_primaries_blue_y
• white_point_x, white_point_y

Mastering Display Luminance:

• max_display_mastering_luminance (nits) - Peak luminance of mastering display
• min_display_mastering_luminance (nits) - Minimum luminance of mastering display

Content Light Levels:

• max_content_light_level (nits) - Maximum luminance of any pixel (MaxCLL)
• max_frame_average_light_level (nits) - Maximum average luminance of any frame (MaxFALL)

Important Notes:

1. Simplified API: With display_static_frame(), HDR metadata and matrix are set in a single call
2. Low-level API call order: When using the low-level API, set_hdr_metadata() must be called before setup_output()
3. Frame-level metadata: Metadata is embedded in every video frame, not set globally
4. Matrix consistency: When using the simplified API, the same matrix parameter is used for both metadata and R'G'B' →Y'CbCr conversion. With the low-level API, ensure consistency between set_hdr_metadata() and conversion functions.
5. Transfer function: The library only sets the metadata - you must apply the actual transfer function (PQ / HLG curve) to your R'G'B' data before conversion
6. All 14 metadata fields supported: The library implements all SMPTE ST 2086 / CEA-861.3 HDR metadata fields including display primaries, white point, mastering display luminance, and content light levels
7. Matrix / Resolution restrictions:
   • Rec.601 is only supported for SD display modes (NTSC, PAL, etc.) and is the only matrix supported for SD
   • Rec.709 and Rec.2020 are only supported for HD and higher resolutions (720p, 1080p, 2K, 4K, 8K, etc.)
8. HLG static metadata is suppressed on transmit by the Blackmagic SDK: When EOTF is HLG, the SDK transmits zero values for all HDR static metadata fields (display primaries, white point, mastering display luminance, MaxCLL, MaxFALL) over HDMI, even when populated explicitly via set_hdr_static_metadata(). The receive side reads non-zero values faithfully when present in an incoming HLG signal, so the suppression is transmit-side, not receive-side. SDI conveys HDR static metadata via SMPTE ST 2108 ANC packets (separate from VPID, which carries only EOTF and matrix); the S2108 path is used for PQ. For HLG on SDI, the static metadata is not present in the captured signal. This is consistent with the view that HLG is display-referred but stricter than CTA-861.3 permits. See HDMI Input Notes for details.

HDMI Input Notes

This section documents behaviour observed on tested hardware (DeckLink UltraStudio 4K Mini) for HDMI input capture. SDI input is generally more deterministic; the notes below apply specifically to the HDMI path.

EDID is partially controllable

The permutations of EDID negotiation between source and sink mean there is always some degree of uncertainty with HDMI capture; you should test your own setup to confirm behaviour.

The DeckLink HDMI input advertises an EDID to source devices. The library writes the dynamic-range advertisement to enable SDR + HDR PQ + HLG (the SDK default omits HLG). Other EDID parameters are not exposed through the SDK and are therefore not available in this library.

Format detection at cold start

Connect your source before starting capture for reliable format detection. The SDK's wire-format auto-detection runs cleanly at the start of a capture session and identifies the input as Y'CbCr or R'G'B' based on the source's HDMI signalling.

Some mid-stream source switches are not detected

Most wire-format changes during an active capture — Y'CbCr ↔ R'G'B' transitions within the HDMI signalling protocol — are detected correctly by the SDK and the capture format updates accordingly.

The exception is the HDMI ↔ DVI protocol switching exposed in the AJA Control Panel for AJA devices. On tested hardware the HDMI → DVI direction is not reliably detected — the SDK continues to interpret incoming bytes in the previous format until capture is restarted. The reverse direction (DVI → HDMI) is detected correctly. If you switch source protocol mid-capture, stop and restart capture.

8-bit R'G'B' sources arrive as 10-bit R'G'B' with LSB padding

Sources sending 8-bit R'G'B' over HDMI (DVI computer signals, AJA devices configured for DVI output in the AJA Control Panel, etc.) arrive as 10-bit R'G'B' with the source's 8-bit values in the high 8 bits and zero LSBs. No information is lost; this is a faithful representation.

The maximum reachable 10-bit value from an 8-bit source is 1020 (= 255 × 4), not 1023. Naively treating LSB-padded 8-bit content as native 10-bit (dividing by 1023) gives values systematically ~0.3% lower than the equivalent 8-bit code divided by 255 — at every non-zero point, not just at peak. (x<<2)/1023 ≠ x/255 for any x except zero; the two denominators are just incompatible.

When the user requests pixel_format=PixelFormat.BGRA at start_capture, the library handles this automatically: each channel is right-shifted by 2 before float conversion (÷ 255) or uint8 extraction, so capture_frame_as_rgb and capture_frame_as_uint8 recover the exact 0.0–1.0 (or 0–255) mapping for content originating from 8-bit.

BGRA capture and the hardware conversion assumption

When capturing with pixel_format=PixelFormat.BGRA, the SDK on tested hardware delivers actual BGRA frames only when the source is Y'CbCr — performing the matrix conversion and range expansion in hardware, yielding full-range 8-bit BGRA. (R'G'B' sources are delivered as 10-bit R'G'B' regardless of what was requested; see the previous section for how the library handles that case.) The library therefore treats BGRA-delivered bytes as full range and applies any input_narrow_range-related conversion accordingly (input_narrow_range describes the wire signal, not the bytes the library receives). This matches observed behaviour on tested hardware but isn't a documented SDK guarantee; if a future driver or hardware delivered narrow-range BGRA from a Y'CbCr source, output from capture_frame_as_uint8 / capture_frame_as_rgb would be slightly compressed.

Caveat — full-range Y'CbCr sources: behaviour unverified. The SDK's hardware Y'CbCr → BGRA conversion has been observed to assume narrow range (64-940 luma in 10-bit mapped to 0-255 BGRA) with the conventional broadcast Y'CbCr sources we tested. Whether the SDK reads the HDMI AVI InfoFrame's YCC Quantization Range (YQ) field and handles correctly-flagged full-range Y'CbCr sources (e.g. computer GPU output configured for Y'CbCr-Full, certain pattern generators) differently is untested and unknown at this time. If YQ is ignored, full-range Y'CbCr values outside 64-940 would be clipped in hardware (0-63 → 0, 941-1023 → 255) before the library sees the frame; no software-side input_narrow_range=False could recover those values. To be safe with potentially full-range Y'CbCr sources, capture as PixelFormat.YUV10 or PixelFormat.YUV8 instead of PixelFormat.BGRA and convert with input_narrow_range=False at the library level — that path preserves the original code values regardless of how the hardware interprets YQ.

Caveat — narrow-range sub-blacks and super-whites are lost in BGRA. Narrow-range Y'CbCr legitimately carries codes below 64 (sub-blacks, footroom) and above 940 (super-whites, headroom). The SDK's narrow-assumed Y'CbCr → BGRA conversion clips these to 0 or 255 at the BGRA output. If you then request output_narrow_range=True to recover an 8-bit R'G'B' representation, the nominal 16-235 range round-trips faithfully but the sub-black region (codes 0-15) and super-white region (236-255) will be empty — that signal was lost in hardware. Less serious than the full-range case (nominal content is preserved), but if you need to retain footroom/headroom information, capture as PixelFormat.YUV10 or PixelFormat.YUV8 and convert in software.

Y'CbCr → BGRA hardware conversion honours signalled matrix metadata

When capturing as BGRA from a Y'CbCr source, the SDK's hardware Y'CbCr → R'G'B' conversion uses the matrix coefficients (Rec.601, Rec.709, or Rec.2020) signalled in the source's frame metadata. Verified via SDI and HDMI loopback with Rec.709 and Rec.2020 sources. Applications working with BGRA captures don't need to apply their own matrix conversion.

HLG static metadata: faithful on receive, suppressed on Blackmagic transmit

The Blackmagic SDK reports HDR static metadata faithfully from HLG sources over HDMI. When an HLG source transmits a populated HDR Static Metadata InfoFrame — Rec.2020 primaries, D65 white point, mastering display luminance — those values arrive on the capture side intact. MaxCLL and MaxFALL may legitimately be zero for HLG content even when the other fields are populated, since HLG is display-referred. If you need mastering display information from an HLG source, the receive side will not strip it.

However, when this library — or anything else built on the Blackmagic SDK — transmits HLG over HDMI, the SDK suppresses HDR static metadata. Even populating an HdrStaticMetadata struct explicitly via set_hdr_static_metadata() produces zero values on the wire, as confirmed by Blackmagic-to-Blackmagic loopback. If you capture HLG and the source is a Blackmagic-SDK-based transmitter, you will see all-zero static metadata. The behaviour is consistent with the view that HLG is display-referred and does not require mastering display information at the receiver, but it is stricter than CTA-861.3 permits.

SDI uses a different mechanism: HDR static metadata is conveyed via SMPTE ST 2108 ANC packets, separate from VPID (which carries only colorimetric descriptors like EOTF and matrix). The S2108 path is used for PQ. For HLG on SDI, the static metadata is not present in the captured signal — only EOTF and matrix come through.

Troubleshooting

Common Issues

"DeckLink output module not found"

• Build and install the C++ extension: pip install -e .
• Check that pybind11 is installed: pip install pybind11

"Could not create DeckLink iterator"

• Install Blackmagic Desktop Video software
• Ensure DeckLink device is connected and recognized by the system
• Check device drivers are properly installed

"Could not find DeckLink device"

• Verify device is connected and powered
• Check device appears in Blackmagic software (Media Express, etc.)
• Try different device index: output.initialize(device_index=1)

Build errors about missing headers

• The SDK headers are included in the repository under _vendor/decklink_sdk/
• If you need to use a different SDK version, update the paths in CMakeLists.txt
• On Linux, ensure headers are accessible to the build system

Permission errors (Linux)

• Add user to appropriate groups: sudo usermod -a -G video $USER
• Log out and back in for group changes to take effect

HDR output not displaying correctly

• Simplified API: Specify both matrix and hdr_metadata in display_static_frame() - they're automatically set correctly
• Low-level API: Call set_hdr_metadata() BEFORE setup_output() - metadata is embedded in each frame
• Ensure matrix consistency: same value in both metadata and R'G'B' →Y'CbCr conversion

Testing Your Installation

Run the example script to test your installation:

python example_usage.py

This will show available devices and let you test various output modes.

Tools

pixel_reader

The pixel_reader tool captures and analyses video input from a DeckLink device, displaying pixel values and metadata. This is useful for verifying output from the library by looping a DeckLink output back to its own input.

Build:

cd tools
make

Usage:

./pixel_reader [device_index]

See tools/README.md for more detail.

The tool displays:

• Pixel format and color space (R'G'B' 4:4:4, Y'CbCr 4:2:2, etc.)
• Resolution and frame rate
• Metadata: EOTF (SDR / PQ / HLG), matrix (Rec.601 / Rec.709 / Rec.2020)
• Pixel values at selected coordinates in native format (code values)

Use this tool to verify that matrix and EOTF metadata are being set correctly by the output library.

Platform-Specific Notes

Windows

• Requires Visual Studio Build Tools or Visual Studio with C++ support
• DeckLink SDK typically installs to Program Files

macOS

• Requires Xcode Command Line Tools
• May need to codesign the built extension for some versions

Linux

• Requires build-essential package
• May need to configure udev rules for device access
• Some distributions require additional video group membership

Contributing

1. Fork the repository
2. Create a feature branch
3. Make your changes
4. Add tests if applicable
5. Submit a pull request

License

This project is licensed under the MIT License - see the LICENSE file for details.

Blackmagic DeckLink SDK License

This repository includes header files from the Blackmagic DeckLink SDK v14.1. These header files are redistributable under the terms of the Blackmagic DeckLink SDK End User License Agreement (Section 0.1), which specifically exempts the Include folder headers from the more restrictive licensing terms that apply to other parts of the SDK.

Important notes about the SDK headers:

• The header files in _vendor/decklink_sdk/{Mac,Win,Linux}/include/ directories are from the Blackmagic DeckLink SDK
• These headers are required only for building the library from source
• Runtime usage requires the Blackmagic Desktop Video software to be installed separately
• The SDK headers are provided under Blackmagic Design's EULA - see _vendor/Blackmagic Design EULA.pdf for full terms
• Download the complete SDK and Desktop Video software from: https://www.blackmagicdesign.com/developer

The Blackmagic DeckLink SDK is © Blackmagic Design Pty. Ltd. All rights reserved.

Support

• Check the Issues page for known problems
• Review Blackmagic's official DeckLink SDK documentation
• Ensure your DeckLink device is supported by the SDK version

Acknowledgments

• Blackmagic Design for the DeckLink SDK
• pybind11 project for the C++/Python bindings
• Contributors and testers
• Special thanks to Zach Lewis and Gino Bollaert