About this project

Great video quality is a fundamental part of a great live stream experience. A crystal clear image is essential to bring your unique vision to your audience in an immersive way.

Our team understands this only too well, which is why a few months ago we gave ourselves the ambitious goal of making major improvements to the quality of the video content that is recorded with or broadcasted through StreamYard. This way, we hope to bring both creators and viewers the best possible streaming experience on our platform.

We're thrilled to share that after several weeks of hard work, we've achieved our goal: The quality of StreamYard broadcasts is now significantly higher across the board.

Here's what you'll notice going forward:

  • Clearer details and text that remain sharp even in lower lighting
  • Smoother motion with significantly less stuttering during dynamic scenes
  • Better color accuracy that makes your content more vibrant and engaging

To accomplish this, the team combined scientific tools with real-world testing:

  • Adopted VMAF to measure video quality, the same professional-grade measurement standard Netflix first implemented to perfect their streaming stack.
  • Comprehensive real-world testing. Bright podcast studios, dimly lit environments, highly dynamic crowded scenes, and everything in between.
  • A/B testing with real viewers. Testing was important to confirm our improvements had visually noticeable impact on StreamYard’s output.

We go into all the details below—we hope it's an interesting read for anyone curious about how we approach quality engineering, from identifying tools and methodology to measuring the outcome of our interventions.

And, of course, a huge thank you to our engineering team for all the effort poured into this project. 💙

Creating a realistic test video

As a starting point, the video engineering team designed a composite test video that captures several real-world streaming scenarios:

  • Webcams under good lighting conditions (e.g., studio or podcast setups)
  • Webcams under low lighting conditions (typical home or dim environments)
  • Crowded scenes (to emulate busy in-person events)
  • Professional setups (to validate performance for advanced broadcasters)

This mix allowed the team to stress-test many common use cases and verify that improvements in one scenario didn't degrade quality in another. All clips were then stitched into a single reference asset, used consistently across experiments.

To support robust analysis, the team embedded two simple but powerful instruments directly into the video: a frame counter and a color marker. These make it much easier to synchronize, inspect, and debug results.

🧮 Frame Counter One of the hardest parts of video-quality evaluation is making sure that the original and processed videos are perfectly aligned. Even a one-frame offset can distort metrics such as VMAF or make manual comparison misleading. This is where a frame counter comes in handy.

How a frame counter works:

  • At the beginning of the test video, a short sequence of black frames with white numbers appears in the center of the screen — 1, 2, 3, and so on.
  • The frame counter gives every frame a unique visual ID, which makes alignment trivial.
  • In automated analysis, scripts read the frame number directly from each frame. This enables frame-accurate synchronization of source and processed videos, ensuring that VMAF always compares the intended pair of frames.
  • In manual review, using tools like DaVinci Resolve, reviewers can place the videos side by side and scrub frame by frame. If both videos show “Frame 137” at the same time, alignment is guaranteed.

This approach enables precise, frame-level inspection and greatly increases confidence in any visual differences that are observed.

🎨 Color Marker Subtle playback issues such as dropped or duplicated frames are notoriously hard to spot by eye. They can create stutter or uneven motion even when the underlying encoder and network are behaving correctly. That's where a color marker can help.

How a color marker works:

  • A small colored box in the corner of the video cycles through colors in a fixed sequence (red → green → blue → … → red again), repeating throughout the entire clip.
  • The color marker acts as a lightweight fingerprint for each frame, which should display the next color in the predefined sequence.
  • If two consecutive frames show the same color, a frame was repeated, indicating potential stutter. If a color in the sequence is skipped, a frame was dropped, which results in choppiness.

Together, the frame counter and color marker give both scripts and reviewers reliable tools to detect timing issues, synchronize assets, and understand exactly what is happening at the frame level.

Measuring video quality with VMAF

For objective evaluation, the team adopted VMAF (Video Multi-Method Assessment Fusion), an open-source metric developed by Netflix to estimate perceived video quality.

The measurement loop looked like this:

  1. Start a StreamYard session using a virtualized webcam as the video source.
  2. Record the resulting VOD (Video on Demand) produced by StreamYard.
  3. Compare the output video (VOD) with the original reference video using VMAF.

VMAF outputs a score representing how close the processed video appears to the original: higher scores correspond to better perceived quality. This provided a consistent, repeatable way to quantify the impact of each change in the pipeline.

Mimicking a real webcam To evaluate how StreamYard behaves with live webcam input, the team used OBS (Open Broadcaster Software). OBS provides a virtual webcam feature, which can feed the test video into StreamYard as if it were a physical camera.

This setup mirrors how creators typically stream, but with full control and repeatability: the same test content can be replayed on demand, without re-recording footage every time.

Identifying bottlenecks and fine-tuning

To understand where quality was being lost, the engineering team deconstructed the StreamYard video pipeline into its major components and measured each step independently.

At a high level, StreamYard’s architecture operates as follows:

  1. The client (the creator’s browser) sends webcam video to a server known as an SFU (Selective Forwarding Unit).
  2. The SFU distributes that video to other participants in the studio.
  3. In parallel, various backend services handle recording (to generate the VOD) and forwarding to external destinations (YouTube, Facebook, and others).

To locate quality degradation, the team recorded video directly at the SFU, measuring the impact of capture, encoding, and network conditions up to that point. They then bisected the pipeline, meaning they introduced measurement points at intermediate stages (for example, directly at the SFU, then later in the backend) and compared VMAF scores between them.

By always checking a midpoint first, they could quickly narrow down which section of the pipeline introduced the largest quality drop and then zoom in on that specific segment.

Once the key pressure points were identified, the team iterated on a range of parameters to make the resulting video sharper, smoother, and more stable. This included:

  • Bitrate allocation at different stages of the pipeline
  • Backend tuning around encoding and streaming behavior
  • Additional optimizations to balance sharpness, motion smoothness, and overall reliability

Each configuration was validated using the test video, VMAF scores, and visual inspection tools, gradually converging on settings that delivered a consistently better experience across different scenarios and network conditions.

Running manual A/B tests

While automated tools like VMAF give us objective quality scores, we also wanted to understand how real people perceived the difference in quality, especially since small technical improvements don’t always translate to a visibly better experience.

To test this, we ran A/B survey tests with about one hundred users with different video expertise, showing them videos from different setups as unlabeled short clips in a random order and asking which one looked better. Each clip represented a specific scenario (e.g., good lighting, low lighting, crowd movement, etc.), allowing us to evaluate each use case individually.

This approach turned out to be a very powerful decision-making tool, often more reliable than VMAF results alone, especially since automated metrics can sometimes be misleading or overly sensitive to small variations.

By combining quantitative analysis (VMAF) with human feedback (A/B testing), we made sure our improvements not only looked better on paper but also felt better to real viewers.

What this means for your streams

This work serves a single purpose: making your streams look closer to what you imagined when you hit “Go live”.

Concretely, these changes translate into:

  • Sharper details and text, even in low lighting conditions, or when the stream has added elements such as lower-thirds and overlays.
  • Smoother motion, with fewer visible stutters during camera moves, hand gestures, or crowd movement.
  • More accurate and stable colors across a wide range of lighting setups.

The result for creators and their audiences is a streaming experience that offers higher video fidelity as StreamYard keeps investing in video quality as an ongoing commitment.

Optimizing for video quality
StreamYard’s new video pipeline does a lot of heavy lifting behind the scenes, but your setup matters. A reliable connection, good lighting, and a great webcam can make these improvements even more visible, especially for important broadcasts like launches, webinars, or live events. For tips on how to test and improve your network quality, you can check out our Help Center.

Comece a criar com o StreamYard ainda hoje

Comece já: é grátis!