Rendering at higher resolution than needed, then downscaling — kills aliasing and edge flicker. Standard for high-frequency patterns and fine geometry.
If you're working with high-frequency patterns — fine grids, textile details, wire mesh — oversampling is unavoidable. You don't render your 3D scene at the desired output resolution, but at a multiple of it — typically 2x, 4x, sometimes 8x — and then scale the result back down. The result: aliasing artifacts and annoying edge shimmering disappear because the sampling rate is below the critical Nyquist frequency, thus suppressing moiré effects.
On set or in the edit, you'll quickly notice the problem: a fine net in the background starts to shimmer, black hair against bright light shows jaggies, or a jacket with a fine pattern dances with the slightest camera movement. Oversampling is then your first choice. The computational effort is considerable — 4x means a 16x increase in pixel output computationally — but that's the cost-benefit calculation in VFX: a night of render time saves a lot of compositor work and frame-by-frame stabilization.
In practice, you distinguish between two scenarios: Firstly, the classic anti-aliasing strategy during rendering: ray tracers and scanline renderers offer integrated oversampling (jitter sampling, adaptive sampling). This is relatively inexpensive and often the standard. Secondly, downstream upsampling — you deliberately render oversized and then compress — this works but requires significantly more memory in the editing system and is less elegant. The first method is always preferable in 3D software.
An important aspect: oversampling also helps with motion artifacts. When an object moves very quickly through the frame or the camera pans, higher sampling resolution stabilizes temporal continuity. This is particularly relevant for CGI vehicles or fast cuts in action sequences. You'll find that the finished element runs smoother and is less prone to "flagging" — an invisible but crucial difference between ambitious VFX and cheap compositing work. Adaptive oversampling, where only critical areas are ramped up, saves computation time — modern renderers now work this way by default.