ofxNozzle

This codebase is currently in its AI-slob prototyping phase: the code runs on momentum, vibes, and plausible intent.
Proper debugging will be introduced once demand graduates from hypothetical to measurable.

openFrameworks addon for GPU texture sharing via nozzle — a cross-platform alternative to Syphon (macOS) and Spout (Windows).

Shares textures between processes on the same machine using platform-native GPU interop: Metal/IOSurface on macOS, D3D11 on Windows, DMA-BUF on Linux.

Disclaimer / Notice

This library is currently a work in progress and contains many incomplete features and unverified implementations. Although it may appear usable at first glance, it may not function correctly.

Please use it with the understanding that no guarantees are made regarding its behavior, and perform debugging, validation, and review as needed. If you encounter problems, please do not become angry; instead, contributions in the form of Issues or Pull Requests would be greatly appreciated.

Platform Support

Platform	Backend	Status
macOS (Apple Silicon)	Metal/IOSurface	Supported
macOS (Intel)	Metal/IOSurface	Supported
Windows	D3D11	Supported
Linux	DMA-BUF	Supported

Setup

Clone into your addons/ directory:

cd <your_oF_project>/addons
git clone --recurse-submodules https://github.com/nozzle-io/ofxNozzle.git

The nozzle static library is bundled in libs/nozzle/. No separate build needed.

Usage

Sender

#include "ofxNozzle.h"

class ofApp : public ofBaseApp {
    ofFbo fbo;
    ofxNozzleSender sender;

    void setup() override {
        fbo.allocate(1920, 1080, GL_RGBA);
        sender.setup("myTextureStream");
    }

    void draw() override {
        fbo.begin();
        ofBackground(0);
        // draw your content here
        fbo.end();
        sender.publishTexture(fbo.getTexture());
    }
};

Receiver

#include "ofxNozzle.h"

class ofApp : public ofBaseApp {
    ofxNozzleReceiver receiver;

    void setup() override {
        receiver.setup("myTextureStream");
    }

    void draw() override {
        if (receiver.receive()) {
            receiver.draw(0, 0);
        }
    }
};

API Reference

ofxNozzleSender

Method	Description
setup(name)	Initialize sender with a name
close()	Release resources
publishTexture(const ofTexture&)	Publish ofTexture to shared texture
publishTexture(GLuint, GLenum, int, int, int)	Publish raw GL texture
setMetadata(key, value)	Attach metadata to published frames
isSetup()	Check if initialized
getName()	Get sender name

ofxNozzleReceiver

Method	Description
setup(name, timeoutMs)	Connect to sender by name. Default timeout: 0 (non-blocking)
close()	Release resources
receive()	Poll for new frame. Returns true if frame available
getTexture()	Get the received texture as ofTexture&
draw(x, y, w, h)	Draw received texture at position/size
draw(x, y)	Draw at original size
isConnected()	Check if sender is alive
getSenderName()	Get connected sender's name

Supported GL Formats

Sender GL Formats

The sender supports 14 GL internal formats:

GL Internal Format	nozzle format
GL_R8	r8_unorm
GL_RG8	rg8_unorm
GL_RGBA8	rgba8_unorm
GL_BGRA8_EXT	bgra8_unorm
GL_SRGB8_ALPHA8	rgba8_srgb
GL_R16	r16_unorm
GL_RG16	rg16_unorm
GL_RGBA16	rgba16_unorm
GL_R16F	r16_float
GL_RG16F	rg16_float
GL_RGBA16F	rgba16_float
GL_R32F	r32_float
GL_RG32F	rg32_float
GL_RGBA32F	rgba32_float
GL_R32UI	r32_uint
GL_RGBA32UI	rgba32_uint

Unsized formats (GL_RGBA, GL_BGRA, GL_RGB) and 3-channel formats (GL_RGB8) are normalized to their 4-channel RGBA equivalents automatically.

Receiver GL Formats

The receiver maps nozzle formats to platform GL texture parameters. On Linux the normal receiver path binds the EGLImage imported by nozzle core from DMA-BUF metadata; the CPU upload path is an explicit fallback only:

nozzle format	GL Internal Format	GL Format	GL Type	Notes
r8_unorm	GL_R8	GL_RED	GL_UNSIGNED_BYTE
rg8_unorm	GL_RG8	GL_RG	GL_UNSIGNED_BYTE
rgba8_unorm	GL_RGBA8	GL_BGRA	GL_UNSIGNED_INT_8_8_8_8_REV	Nozzle converts to BGRA internally
bgra8_unorm	GL_RGBA8	GL_BGRA	GL_UNSIGNED_INT_8_8_8_8_REV
rgba8_srgb	GL_RGBA8	GL_BGRA	GL_UNSIGNED_INT_8_8_8_8_REV	CGL sRGB fallback to unorm
bgra8_srgb	GL_RGBA8	GL_BGRA	GL_UNSIGNED_INT_8_8_8_8_REV	CGL sRGB fallback to unorm
r16_unorm	GL_R16	GL_RED	GL_UNSIGNED_SHORT
rg16_unorm	GL_RG16	GL_RG	GL_UNSIGNED_SHORT
rgba16_unorm	GL_RGBA16F	GL_RGBA	GL_HALF_FLOAT	CGL has no GL_RGBA16, falls back to 16F
r16_float	GL_R16F	GL_RED	GL_HALF_FLOAT
rg16_float	GL_RG16F	GL_RG	GL_HALF_FLOAT
rgba16_float	GL_RGBA16F	GL_RGBA	GL_HALF_FLOAT
r32_float	GL_R32F	GL_RED	GL_FLOAT
rg32_float	GL_RG32F	GL_RG	GL_FLOAT
rgba32_float	GL_RGBA32F	GL_RGBA	GL_FLOAT
r32_uint	GL_R32UI	GL_RED_INTEGER	GL_UNSIGNED_INT
rgba32_uint	GL_RGBA32UI	GL_RGBA_INTEGER	GL_UNSIGNED_INT
depth32_float	GL_R32F	GL_RED	GL_FLOAT	Single channel fallback

Integer textures require a custom shader with usampler2D / texelFetch() for correct sampling. The default openFrameworks shader will not render uint data correctly.

Architecture

All Objective-C types are hidden behind pimpl. Headers are pure C++.

Sender flow:
  User's ofTexture → publishTexture() → nozzle core → shared state

Receiver flow:
  macOS:   nozzle acquire_frame → IOSurface → CGLTexImageIOSurface2D → cached GL texture → ofTexture
  Windows: nozzle acquire_frame → D3D11 texture → glTexSubImage2D → cached GL texture → ofTexture
  Linux:   nozzle acquire_frame → DMA-BUF EGLImage → glEGLImageTargetTexture2DOES → cached GL texture → ofTexture
           fallback only: DMA-BUF mmap → glTexSubImage2D; logs as CPU fallback and ignores modifiers

Requirements

• openFrameworks 0.12+
• macOS 12.0+ (Metal framework)
• Windows 10+ (D3D11)
• Linux (DRM/KMS, DMA-BUF)

License

MIT

Third-party dependencies:

• nozzle — MIT