Nodes/ComfyUI_Gear/Gear · LogC4 Decode + Save EXR
ComfyUI Node

Gear · LogC4 Decode + Save EXR

LogC3's taller sibling — about three extra stops of highlight headroom

By oumad·Created 4 months ago·Updated 3 days ago· 0
Gear · LogC4 Decode + Save EXR
  • image
  • hdr_linear
  • tonemapped_preview
  • exr_paths
filename_prefixgear_hdr/%year%%month%%day%_%hour%%minute%%second%
preview_ev0.0
save_exrtrue

This is the same node as Gear · LogC3 Decode + Save EXR, with a different curve and a different LoRA family to feed it. LogC4 is ARRI's ALEXA 35-era successor to LogC3, and it's wider: where LogC3 caps out around 55 linear (~8.3 stops above mid-gray), LogC4 goes to ~470 linear (~11.3 stops). That's roughly three extra stops of highlight headroom riding in the same [0,1] output - which only matters if your HDR LoRA was actually trained on LogC4 targets.

That family is the V10 generation: v10_logc4_*, klein4b_v10_logc4_*, klein9b_v10_logc4_*. If that's what you're running, this is the decoder you want - the LogC3 node would compress all that highlight range back down and defeat the point of the model.

The inputs and outputs

Identical to its LogC3 sibling, so I'll be brief:

  • image - the [0,1] LogC4 output of your HDR LoRA.
  • filename_prefix - ComfyUI tokens supported (%year%, %month%, %day%, %hour%, %minute%, %second%, %batch_num%); files land in ComfyUI/output/gear_hdr/ and auto-counter so nothing overwrites.
  • preview_ev - exposure shift for the Reinhard graph thumbnail only. The EXR is untouched.
  • save_exr - toggle disk writes off if you just want the in-graph hdr_linear.

Outputs are hdr_linear (scene-linear, up to ~470), tonemapped_preview, and exr_paths. Full workflow, install steps, and troubleshooting live in the LogC3 article - same pack, same install, same gotchas.

The one thing that actually matters

Curve matching. LogC3 and LogC4 output look nearly identical on a monitor, which makes the mismatch easy to miss and harder to spot. If you feed LogC3-era content (LumiVid V9, klein_step*) into this node, you get an EXR that looks fine and is numerically wrong. Read your LoRA's card before you decide which decode node to use - that's the whole game with this pack.

The mechanism is worth a nod though: LogC4's constants come straight from the ARRI white paper, implemented as a pure per-channel torch curve in gear/logc4.py with no OCIO dependency. That's the pack's design in miniature - real, reference-accurate color math with nothing heavy bolted on.

CategoryGear/HDR

Inputs (4)

NameTypeDefaultDescription
imageIMAGE
filename_prefixSTRINGgear_hdr/%year%%month%%day%_%hour%%minute%%second%
preview_evFLOAT0.0-8–8
save_exrBOOLEANtrue

Outputs (3)

NameTypeDescription
hdr_linearIMAGE
tonemapped_previewIMAGE
exr_pathsSTRING