Reducing color depth or resolution to fewer bit levels — 10-bit to 8-bit, 4K to 2K. Required for codec compatibility and file size, but sacrifices tonal gradation.
You reduce the digital information of an image to less storage space or fewer color levels — that is quantization. In everyday VFX, this happens constantly: a 10-bit RAW file from set needs to be down-converted to 8-bit for online editing. A 4K DCP is quantized to HD for social media. Or your color correction project uses 16-bit floating-point, but the final export ends up as 8-bit integer. Each time, you gradually lose color information — and that is precisely the core problem that needs to be managed.
The practical reason for quantization is simple: compatibility and file size. A 10-bit Log image simply has more data than an 8-bit linear image. If your delivery codec (H.264, ProRes, DNxHD) only supports 8-bit, you have to quantize. The same applies to resolution reduction — from DCI-4K to 2K, you halve the pixels, thus you are spatially quantizing. In the grading suite, you notice this immediately: when you work with 10-bit Log masters against 8-bit Rec.709 proxies, the proxies are already quantized. The highlight details that are still present in the master are already missing in the proxy. This forces you to always go back to the uncompressed original master for final color decisions later.
When does it become critical? With multiple quantization steps. If you import an 8-bit file, perform color correction on it (usually 16-bit internally), and then export back to 8-bit, you multiply the quantization error — creating so-called banding. Subtle color gradients turn into visible stripes. That's why in post-production, one works with a *higher* bit depth and only quantizes at the very end. In VFX compositing, this is even more critical: if your elements come from different sources (8-bit stock footage, 16-bit render), you must bring everything into a consistent bit space before you composite. Otherwise, visible artifacts will appear at the transitions.
A rule of thumb from set to edit: quantize as late as possible, as little as possible. Always work internally with the highest available bit depth. RAW remains RAW as long as the workflow allows. Only when the cut is locked and the color grade is final do you move to the delivery format. And even then — if there are critical scenes — you can still keep 10-bit versions for DCP or streaming masters and only quantize to 8-bit for the web. File size is no longer the real problem today. Data loss is.