Considerations for replacing Filmic Tonemapping Curve

Hello Epic,

We are looking into the possibility of replacing the default Filmic tonemapping curve with a custom one, such as GT Uchimura, in order to achieve a different look and color response. The goal would be to keep the existing tonemapper pipeline as intact as possible, and replace only the curve itself. Ideally, features such as bloom, auto exposure, color grading, output device handling, and HDR support would continue to behave as expected.

At the moment, the approach being considered would be to modify the tonemapping implementation directly (for example in PostProcessTonemap.usf or TonemapCommon.usf), rather than apply an additional color transform such as OpenColorIO, since the objective is to replace the tonemapping operator itself.

Since this would need to ship on PC as well as multiple consoles (such as PS5, XSX, and Switch 2), we would like to better understand any potential implications before moving forward.

So, questions are:

1) Are there any known limitations, hidden dependencies, or platform-specific considerations when replacing the tonemapping curve?

2) More generally, is the Filmic curve treated as a self-contained component, or are there parts of the rendering pipeline that rely on its specific behavior?

Any guidance or recommendations would be greatly appreciated

Thank you!

Hi Jon,

If you already preserve a engine fork, modifying the engine’s Filmic path is fair game, and should not be too concerning. The tonecurve function is generic and platform-independent. The primary considerations I would factor in are:

  • Native HDR support: The filmic SDR path diverges from our standard ACES HDR output transforms. Note that 5.8 has experimental support for Filmic + Inverse ACES 2.0 + Forward ACES 2.0 (HDR output) to support “Filmic” in HDR. This has already been significantly improved in Main however.
  • Minimal maintenance burden (for future engine versions)
  • Our shaders have a filmic inverse function but I would not recommend relying on it as it’s quite innacurate. The filmic path has no analytical inverse. ACES 2.0 inverse is very reliable, but not exposed for general use in 5.8.

One thing to be aware of (coming in Main) is that we have recently added initial support for Look Transforms via ACEScct/AP1 3D LUTs in the PPV, which lets users customize their output transforms in a more integrated & generic manner. For instance, one application would be to emulate the look of other tone curves before/while applying standard ACES 2.0 for SDR/HDR output.

Hope this helps,

Eric

Thank you for the quick response! It does answer our questions. PPV’s Look Transforms seems interesting, we are looking forward to try it