◎ Radiance Grade

Here's the thing most ComfyUI users do when their image "needs a grade": they re-roll it with a prompt tweak, or worse, run it back through the sampler with a low denoise. That's the expensive way to do work a colorist does in a second. Radiance Grade is the LGG node - Lift, Gamma, Gain, plus Offset, Contrast, and Saturation - with per-channel control, running in 32-bit float. It's the piece of a proper color pipeline that was missing from most ComfyUI installs.

The KB's post-processing doc makes the argument plainly: almost none of this is AI, and you should reach for the cheap deterministic primitive before the expensive generative one. This is that primitive. A gamma curve is a gamma curve - but here it's a per-channel gamma with a pivot and a JSON export, which is where it stops being a toy.

How it works

The grade math is standard film-print style: lift is a shadow offset, gamma is a midtone power, gain multiplies highlights, offset adds a global shift. Contrast is applied around a pivot (output = (img - pivot) * contrast + pivot), and saturation is luminance-preserving so cranking it doesn't blow out brightness. The whole thing runs in 32-bit float, which is the pack's whole thesis - standard ComfyUI nodes clamp to 8-bit-ish range, and this one doesn't. The pivot tooltip flags that v2.3.3 pins it to 0.18 for scene-linear correctness, which is the difference between grading an sRGB image and grading actual HDR/linear footage.

Two features put it ahead of a bare LGG math node:

• Presets. Thirteen built-in looks (Cinematic Teal & Orange, Bleach Bypass, Film Noir, Golden Hour, Matrix Green, Horror Desaturated, and more) with a preset_strength blend so you can dip into a look instead of committing. You can also load your own from a JSON file in GRADE_PRESETS format via preset_file.
• Grade matching. Connect a reference_image and the node computes the grade automatically to match that image's color statistics in CIE L*a*b* (mean/std), with match_strength blending between your manual grade and the matched one.

Inputs and outputs that matter

The manual controls are nine per-channel values plus the four globals: lift_r/g/b, gamma_r/g/b, gain_r/g/b, offset_r/g/b, then contrast, pivot, saturation. For a beginner: set preset to a look at low preset_strength, then nudge gain for highlights and lift for shadows. The gamma values sit at 1.0 (no change), gain at 1.0, lift/offset at 0.

Outputs are the real payoff: image (the graded 32-bit result) and grade_info - a JSON string of every applied parameter. Wire that into the pack's ApplyGradeInfo node and you can replay the exact grade onto other shots with a strength control. That's the difference between "I got a nice look" and "I have a reproducible grade."

Installing

cd ComfyUI/custom_nodes
git clone https://github.com/fxtdstudios/radiance.git
cd radiance
pip install -r requirements.txt   # or requirements_windows.txt / requirements_linux.txt / requirements_mac_silicon.txt

Or search Radiance in ComfyUI Manager. Linux needs libopenexr-dev before pip. Restart after.

Common issues

The recurring one, straight from the community and the README: this node cannot recover data that isn't there. If your source is an 8-bit sRGB PNG, grading it in float doesn't create headroom - the pack "cannot recover highlight or shadow detail that was already clipped in an 8-bit source." People expect the HDR magic to fix their blown-out sky and it doesn't; that's a capture problem, not a grade problem. Feed it EXR or float output if you want the full dynamic range. Second gotcha: match_strength defaults to 1, so plugging in a reference image will fully override your manual grade until you dial it back. That's intended, but it surprises people the first time.