Precision within VFX pipeline — how accurately data remains consistent between render engines, compositing, and final delivery. Controls color loss and numerical drift.
Once you're working in a VFX pipeline, you quickly realize: a chain of numbers is only as reliable as its weakest link. Internal Accuracy describes how precisely numerical data — color values, positions, luminance — are preserved across all stages from 3D rendering to final file output. It's not about sharpness or visual quality, but about the mathematical integrity in the background.
In practice, this is what happens: you render a VFX plate in 16-bit linear color space, hand it over to the compositor — who might work in 32-bit float for precision — and later your colorist compresses the material for DCP or streaming. Potential errors arise at each of these transitions. Noise in color information, banding in shadows, minimal shifts in contrast. When you work with integer mathematics (8-bit, 16-bit without float), these errors accumulate — especially visible at extreme grades (extremely dark or bright areas) or when you perform multiple transformations. Floating-point rendering (32-bit or higher) maintains more precision but costs more memory and CPU.
You experience this concretely, for example, when you export a particle sim in Houdini with 16-bit fixed-point values and later import it into Maya — the positions drift. Or: your lighting artist renders passes in 8-bit, but the compositor needs the full luminance for color grading — this cannot be reconstructed. Therefore, reputable pipelines work consistently in at least 16-bit linear or 32-bit float, even if the final delivery is 8-bit. The workflow is: maintain the highest possible precision until the end, then resample for distribution. This is your insurance against surprises at the master DCP turnover.
In the context of look development and color pipeline, Internal Accuracy is closely linked to the concept of color space and bit depth — it's about ensuring that your LUT, your OCIO config, and the rendering engine are based on the same computational accuracy. If your renderer calculates in sRGB, but your compositor expects linear, not only color but also luminance distortions will occur. Therefore, you define pipeline-wide: work linearly, store in 16-bit or higher, and only resample to the distribution specification at the end. This is your insurance against surprises at the master DCP turnover.