Passthrough Latency in AR Porn: The 12ms Threshold Explained
Twelve milliseconds. That's the latency budget where AR / passthrough porn either feels believable or feels uncanny. Below it, the actor stays anchored to your real room when you turn your head. Above it, edges shudder, the composite lags behind your motion, and your brain quietly clocks that something's off — even if you can't articulate why. This piece walks through why 12ms is the number, what eats into the budget on Quest 3 and Vision Pro, and the practical setup changes that keep you on the right side of it.
In this guide
Where the 12ms number comes from
Twelve milliseconds is roughly a single frame at 90fps (11.1ms, but we round). It's also the threshold past which most people consciously or sub-consciously notice a mismatch between head motion and visual response. Below 12ms, the brain accepts the composite as real. Above 12ms, the vestibular system starts flagging the inconsistency — turning your head feels like dragging the actor through molasses.
Sensitivity varies. Some users are flagging mismatch at 8ms. Others won't consciously notice until 20ms. The 12ms target is the conservative middle, and it's what serious AR studios test against.
How the budget gets spent
The 12ms isn't the full pipeline — it's specifically the delta between passthrough refresh and composited-content refresh. The full motion-to-photon chain is longer (Meta quotes ~35ms for Quest 3 raw passthrough, Apple quotes ~12ms for Vision Pro). What matters for AR scenes is that the actor and the room stay synchronised.
The chain inside the player roughly looks like:
- Decode the next video frame from the codec (2-6ms for AV1 hardware decode on Quest 3).
- Composite alpha against the latest passthrough capture (1-3ms).
- Submit to the headset compositor (1-2ms).
- Display panel refresh (1-3ms on LCD, sub-1ms on OLED).
Add it up and a well-tuned pipeline runs 6-10ms inside the player. With passthrough capture and motion prediction handled separately by the headset OS, the actor stays glued to the room. Blow any stage of that budget and the composite shudders.
Hardware contributions per headset
Latency floor depends on the device. Approximate numbers from our testing:
- Quest 3 / Quest 3S: 8-10ms in-player on AV1 at 90fps. Best balance of codec support and decoder speed for the price.
- Quest Pro: Slightly slower codec on older Snapdragon XR2 Gen 1, but eye-tracking enables foveated decode hints that recover the gap. Net similar to Quest 3 on well-supported codecs.
- Pico 4 Ultra: 9-11ms typical. Codec stack is solid; passthrough cameras are slightly noisier than Quest 3, but that's a quality issue not a latency one.
- Vision Pro: Hardware would do 5-7ms in-player. Software doesn't exist to use it. Apple's App Store policy makes the chip's headroom unreachable for adult content.
- PSVR2: Higher latency floor due to colour-passthrough being a 2025 software update rather than designed-in. Around 14-16ms; right at the edge of the believability threshold. Covered in our PSVR2 passthrough piece.
Codec choice and decoder timing
AV1 is the right choice in 2026, with one caveat. The Snapdragon XR2 Gen 2 in Quest 3 has hardware AV1 decode that's 2-3ms faster than HEVC for the same bitrate. AV1 at 90fps on a properly-encoded scene sits comfortably under the latency budget.
The caveat: not every AV1 profile is hardware-accelerated. If a studio encodes with an exotic AV1 profile (10-bit colour depth with non-standard tile grid), the decoder falls back to software, latency triples, and the scene stutters. This is rare but real. Heresphere flags software-decoded scenes in the file info pane; if you see "SW decode," that's why the latency's bad.
HEVC remains the safer fallback for older content. Higher decoder maturity, more predictable timing, slightly higher bitrate for equivalent quality. The codec comparison gets a dedicated piece in our AV1 vs HEVC guide.
Streaming vs local playback
Local files (or SMB shares) keep the latency budget purely on the player and headset. Streaming from a CDN adds buffering — typically 100-300ms ahead of the playhead. That buffer doesn't affect passthrough sync (the player reads ahead and the composite works frame-by-frame), but it makes scrubbing and seeking sluggish. Tap the timeline, wait a second, then the scene resumes.
For best-case latency on high-bitrate AV1 scenes, download once and play locally. Our PC-streaming guide covers the local-network setup with SMB that gets you the best of both worlds.
Practical fixes when latency is too high
If a scene feels laggy or the actor's edges shudder, work down this list:
- Let the headset cool. Sustained passthrough at 90fps puts the Snapdragon XR2 Gen 2 close to thermal limits after 30-40 minutes.
- Check the file info pane for "SW decode." If hardware decode isn't engaging, find a different encoding of the same scene.
- Drop frame rate to 60fps if the source is 60fps native. Interpolation to 90fps eats budget you don't need to spend.
- Restart the player. Heresphere and DeoVR both develop slow leaks over long sessions; a fresh launch resets timing.
- Close background apps on the headset. The Quest 3 will happily run a Steam Link tunnel in the background and eat 2-3ms of compositor budget.
FAQ
Why specifically 12ms?
It's roughly where most people's vestibular system starts to flag a mismatch between head motion and visual feedback. Above 12ms, turning your head feels delayed against the real-room composite — and even if you don't consciously notice, the brain registers it as wrong. Some users are sensitive at 8ms; some don't notice until 20ms. Twelve is the conservative threshold studios target.
Quest 3 passthrough latency — what does it measure?
Meta's own number is around 35ms motion-to-photon for passthrough alone, before any compositing. That sounds bad until you realise the brain's tolerance for static composites is much higher than for head-motion mismatch — what matters is the delta between passthrough refresh and the composited actor's refresh, which the player keeps under the 12ms threshold on well-tuned scenes.
Does Wi-Fi streaming add latency?
For local SMB streaming on Wi-Fi 6E from a wired-PC source, network latency is sub-5ms and doesn't meaningfully affect the composite. Where it matters is internet streaming from a CDN — those can add 100-300ms of buffer, which doesn't break passthrough sync (the player buffers ahead) but does make scrubbing and seeking sluggish.
Is Vision Pro's lower latency a meaningful AR advantage?
On paper yes. Vision Pro's R1 chip is dedicated to sensor fusion and passthrough timing, and motion-to-photon latency is roughly half of Quest 3's. In practice this advantage is wasted because Vision Pro has no native AR porn player. You can't run Heresphere or DeoVR on visionOS, so the hardware win never reaches the use case.
Can I check passthrough latency myself?
Approximately. Wave your hand in front of the headset while running a passthrough scene with a static actor. If your hand visibly leads the composited actor by a frame or two, latency is fine. If the actor's edges shudder when you move your head, latency is too high — usually means the player can't keep up with the codec, or the headset is thermal-throttling.
Related on PassthroughTube
- Frame rates and performance
- AV1 vs HEVC codec comparison
- How AR passthrough works
- All passthrough scenes
Cross-link: For traditional VR (non-passthrough) content from many of the same studios, see our sister site VRTubbies — over 41,000 VR scenes by studio.
