Skip to main content
Inspiring
September 24, 2026
Question

Object Matte tool (4.0) unusable and too slow.

  • September 24, 2026
  • 4 replies
  • 21 views

I am wondering if I an asking too much of this tool. I have 14000 frames of 4K footage. And the client wanted (last minuet) to cut out speaker  in front of some diagrams. Using the normal footage. (obviously without a green screen) Below is a small sample of of the image. You can see there isn’t much difference between between the hair and the shade of the blue curtains. Initially I tried the same (new) tool in premiere 2026, but the darkness of the curtains where still bleeding into the hair. I have now tried to roto  using the the object matte (4.0) in after-effects. It is working quite well however it is just so slow, I’m not sure it will ever render. I am now trying to freeze the Object matte, but, again painfully slow. Hoping to be able to render. In the setting, I have tried switching off some refinement settings and dialling down the search area. Gemmi and Lama4 reckon it can be done. So I am wondering if I am not doing something right? Thank you.

 

 

    4 replies

    Warren Heaton
    Community Expert
    Community Expert
    September 25, 2026

    Hi, ​@Joolsey1973:

     

    For rotoscope work, duration is a key factor when it comes to how quickly a job can be completed.  You didn’t mention frame rate, but a 9 minute 43 second clip at 24fps is around 14,000 frames, and every one of them needs to be tracked and refined.

    Even with the Object Mask tool (the Roto Brush's newer AI-assisted sibling), a clip this long is still a heavy lift. It does a much better job propagating a mask across frames than manual roto, but it's not 'set and forget' — it still needs correction passes wherever the subject moves quickly, overlaps other elements, or where the edge detail gets complicated.

    Hair is the classic problem case: the mask engine is essentially finding a hard-ish edge and tracking it forward, but fine flyaway strands, semi-transparent wisps, and hair blowing or shifting frame-to-frame don't behave like a solid, trackable edge — the AI either clumps the hair into a blob (losing the fine detail) or the edge chatters and drifts. That's why hair usually needs its own second pass with the Refine Edge tool, which handles soft, semi-transparent detail the base mask can't — but that's an additional frame-by-frame pass on top of the primary roto work, not a replacement for it.

    Once the mask and the refined edge both look right, the last step is hitting Freeze, which locks the propagated mask into a static, cached matte across the whole range so it stops recalculating and playback/render stays stable. You only want to hit Freeze once you're happy with the result — any further correction after that means unfreezing, fixing, and re-propagating, so it's the final checkpoint, not a shortcut for the earlier passes.

    So you're really doing two correction passes across the whole runtime, then a lock-in step, before the roto is actually done — none of which gets shorter just because the AI-assist did the first pass faster. On a 9:43 clip, especially with a subject who has loose or detailed hair, that adds up to a multi-week task rather than a quick turnaround.

    Since you only have a short time to do this, talk to the client about doing a few seconds of the speaker in front of the diagrams here and there, then cut to fullscreen shots of the diagrams, and then cut back to the speaker.  You can then focus using the Object Mask Tool and Refine Edge Tool for short segments.

     

     

     - Warren

     

     

    Community Expert
    September 24, 2026

    Could you share your system specifications? This is kind of a demanding task.

    Inspiring
    September 24, 2026

    Thanks for the replay. 

    Nishu Kushwaha
    Community Manager
    Community Manager
    September 25, 2026

    Thanks for sharing the specs, your system looks fine. I think the bottleneck is really just the sheer scale.

    My recommendation would be to break the shot into smaller chunks (500–1000 frames each), matte each segment separately, and queue them as separate renders. That way you can process overnight in stages, and if propagation drifts on one section, you only redo that chunk rather than the entire shot.

    Let me know how it goes.


    Thanks,
    Nishu