This is because in browsers in general the video might have data around it that's unrelated to the video. The injected shader might be doing a transform on the coord to select a potentially non-aligned rectangle of the video frame. The injected shader does a transform that maps the [0, 1]^2 range correctly, but values outside of that range may be sampling outside of the sub-rectangle of the video frame. To prevent this we can specify that the operations always clamp to [0, 1]^2