You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
C920 at 1080p cuts frames short, which #40 turns into stutter: step down to a size the camera delivers whole #45
On 0.4.0 my Logitech C920 records MJPEG 1920 × 1080, which #19 picks correctly. The takes stutter. Since #40 the incomplete frames from #24 no longer show as a colour grid, but every skipped frame holds the one before, and at 1080p the C920 here cuts enough of them short that you see it.
#40 already spots every frame the camera cut short, so Monologue can use that count to pick the mode. It's the same idea as #19, based on the frames the camera actually sends instead of the modes it lists, and it adds no control (#14, #3).
Count frames cut short, in the preview and during a take.
Once 4 arrive within 300 frames (over 1%), hold this camera below the current size: bestFormat takes a maximum area and picks the next size down. Outside a take the camera restarts at once. During a take only the limit is saved, and the next start uses it.
Remember the limit under the camera's ID in the settings, so the next start goes straight to it.
The decision comes as soon as the fourth frame arrives, so at 1080p it takes about 3 s, and a size that delivers keeps running without any wait. With 720p at 0.7%, the threshold leaves it alone.
I've been running this locally on 0.4.0 with #43 (and #44 for the last two takes below). The C920 went 1920 × 1080, then 1600 × 896, then 1280 × 720, learning during my first two takes. Since then:
Take
Size
Frame rate
Gaps
8.3 s
1280 × 720
30.0 fps
0
5.7 s
1280 × 720
30.0 fps
0
5.4 s
1280 × 720
30.0 fps
0
It's 43 lines across 5 files, with a formatSelection case built from the C920's real mode list. All backend tests pass. One thing to know: the camera ID is the device node (/dev/video0), so another camera that later gets the same node inherits the limit.
I can open a PR if this is the direction you want.
Closing this myself. The frame loss at 1080p is real (about 5% of frames on my C920), but I can't see it in a normal talking-head take, so 1080p is fine for me.
What I actually heard as stutter was the audio crackle that #44 fixes. I checked whether the clicks lined up with the cut frames, and they don't: in three 1080p takes the clicks landed near a cut frame no more often than chance would put them there. A 720p take with no cut frames at all still crackled until I applied #44. I've confirmed that on #44.
The step-down patch works, if anyone runs into a camera where the cut frames do show. As it stands, the extra logic isn't worth it.
Activity
RemyDuijkeren commentedon Nov 14, 2026
Closing this myself. The frame loss at 1080p is real (about 5% of frames on my C920), but I can't see it in a normal talking-head take, so 1080p is fine for me.
What I actually heard as stutter was the audio crackle that #44 fixes. I checked whether the clicks lined up with the cut frames, and they don't: in three 1080p takes the clicks landed near a cut frame no more often than chance would put them there. A 720p take with no cut frames at all still crackled until I applied #44. I've confirmed that on #44.
The step-down patch works, if anyone runs into a camera where the cut frames do show. As it stands, the extra logic isn't worth it.