Gemini 3 Flash

Here is a breakdown of how these three points interact and why darktable offers three different ways to set them.

1. The Geometry: Why Three Points?

You are correct that a sigmoid (specifically the AgX implementation, which is a modified log-logistic or similar curve) needs anchors.

  • The 18% Gray Point (The Pivot): In darktable’s AgX, the middle gray point is fixed at 0 EV (input) and 18.45% (output). This is the „anchor of reality.“ Because darktable is scene-referred, we assume that the Exposure module has already been used to „place“ middle gray correctly. Therefore, the curve must pass through (0, 0.18).
  • The Dynamic Range (Black and White): These define the „reach“ of the curve. If you set the white point to +6 EV, you are telling the algorithm: „Map the value that is 6 stops above gray to pure white (1.0).“

2. The Three Approaches: Why the Choice?

The reason you are offered three methods is that „Black“ and „White“ are not absolute physical facts in a RAW file; they are interpretations.

A. The Default (-10 EV / +6.5 EV)

  • What it is: A „one size fits most“ safety net.
  • Why use it: It mimics a standard film-like roll-off. It provides a consistent look across a series of photos. If your scene has 10 stops of dynamic range, this curve will look „normal.“ If your scene only has 4 stops, it will look „flat“ because the curve is stretching to points that don’t exist in your data.

B. The Pickers / Auto-Tune (The Histogram Method)

  • What it is: An analysis of the actual pixels in the current image.
  • Why use it: This is „Technical Optimization.“ It finds the absolute darkest and brightest pixels and sets the anchors there.
  • The Problem: If you have a tiny specular highlight (a reflection on a chrome bumper), the picker will set the white point very high to „save“ that highlight, which might make the rest of the image look too dark or low-contrast. It is data-driven, not aesthetically driven.

C. The Camera Icon (The Exposure-Linked Method)

  • What it is: This is the most misunderstood part. It calculates the dynamic range based on the Exposure module’s settings and the camera’s metadata.
  • Why use it: In a scene-referred workflow, the Exposure module defines your „working space.“ If you increased exposure by +2 EV in the Exposure module to see into the shadows, the „Camera Icon“ in AgX adjusts the anchors to compensate for that shift. It tries to maintain a „neutral“ mapping where the dynamic range of the sensor is mapped proportionally to the display.
  • Kofa’s „vague“ answers: The reason experts are vague is that this depends on the „headroom“ of your specific sensor. It’s an attempt to automate the „Exposure + Tone Mapping“ relationship.

3. How to Solve the „Which one to use?“ Dilemma

Think of AgX not as a „correction“ tool, but as a „Window“ through which you view your high-dynamic-range data.

  1. The Exposure Module is Step 1: You MUST set your 18% gray first using the Exposure module. AgX assumes 0 EV is gray. If your exposure is wrong, AgX will never look right.
  2. Use the Camera Icon if you want a technically „neutral“ starting point that respects how much you’ve pushed the RAW data in the Exposure module.
  3. Use the Pickers if you have a very specific scene (e.g., a low-contrast foggy day) and you want to maximize the use of the 0-1 output range immediately.
  4. Manual Adjustment (The „Boris“ Way) is usually best:
    • Set Exposure for the subject’s face/midtones.
    • Adjust the White Point in AgX until the highlights stop clipping (check the RAW clipping indicator).
    • Adjust the Black Point until the shadows have the desired „depth.“

The confusion stems from the fact that AgX doesn’t „determine“ the 0 EV point—YOU do (in the Exposure module).

  • Exposure Module: Moves the whole world left or right on the EV scale.
  • AgX: Decides how much of that world we can see (White/Black points) and how „bent“ the glass is (the slope/contrast).

The three methods are simply different ways to guess where the „edges“ of your world are. The Camera Icon guesses based on hardware/settings; the Picker guesses based on pixels; the Default doesn’t guess at all.

Fragen zu diesem Thema? Diskutiere mit uns im Forum!

Letzes Update: Februar 4, 2026