Gyroflow: Stabilizing Video With Gyroscope Data, Not Guesswork
You've probably shot some footage that looked great in camera but turned into a shaky mess when you played it back. Most video stabilization tools work by analyzing the image itself—tracking features, guessing motion, and hoping for the best. That approach has limits, especially when the camera is moving fast or the scene lacks visual detail. Gyroflow takes a different route: it uses actual gyroscope data recorded by the camera to stabilize your footage with precision that image-based algorithms can't match.
Gyroflow is an open-source application that stabilizes video using motion data from gyroscopes and accelerometers. Modern action cameras like GoPro, Sony, and Insta360 record this telemetry internally, and Gyroflow can read that data to calculate exactly how the camera moved during every frame. It can also pull gyro data from external sources, like Betaflight blackbox logs from FPV drones. The result is stabilization that understands the physical motion of the camera, not just the visual artifacts of that motion.
What It Does
At its core, Gyroflow takes your video file and the gyroscope data associated with it, then calculates the precise camera orientation for each frame. It applies the inverse of that motion to produce a stable output. The application handles all of this in real-time—you get a live preview while you tweak parameters, and all the heavy lifting runs on your GPU with multi-threaded processing.
Beyond basic stabilization, Gyroflow includes rolling shutter correction, which is critical for cameras with CMOS sensors that read out line by line. It supports full Sony camera metadata, meaning it can read recording parameters, automatically detect the lens, and even work with Sony's internal IBIS and OIS enabled. For GoPro users, it can stabilize footage that was already captured with Hypersmooth turned on (Hero 8 and up), effectively layering Gyroflow's stabilization on top.
The application works with high-bit-depth video natively—10-bit, 16-bit 4:4:4:4, and even 32-bit float for OpenEXR sequences. It processes directly on YUV data to avoid quality loss from color space conversions. It also handles underwater footage by correcting the distortion caused by the housing and water interface.
The tech stack is Rust-based (the project is built in Rust) with GPU compute shaders handling the heavy processing. The gyroscope parsing relies on the telemetry-parser library, which supports a wide range of camera and drone telemetry formats out of the box.
Why It's Cool
The most interesting thing about Gyroflow is that it treats stabilization as a physics problem, not a computer vision problem. That distinction matters.
It works when visual tracking fails. If you're filming a blank wall, a dark scene, or fast motion blur, image-based stabilizers have nothing to lock onto. Gyroflow doesn't care what the image looks like—it knows exactly how the camera moved because the gyro data tells it.