RAW Load

RAW Load is the node that turns Darkroom from "another film filter pack" into a real post-production suite. It decodes actual camera RAW files - .cr3, .nef, .arw, .raf, .dng, .rw2, .orf, and more - through rawpy/LibRaw, with the demosaic algorithm, white balance, highlight handling, output colorspace, and even Adobe .dcp camera looks exposed as inputs. If you shoot real photos and want to grade them with the same nodes you use on generated images, this is the bridge. If you only generate, you can mostly skip it - but it's also the only honest way to check how your film emulation behaves on sensor data instead of diffusion output.

The whole pack is designed around it. All of Darkroom's processing happens in linear light, and RAW Load hands you a genuine linear decode, so the film stocks and print chain actually have the dynamic range to chew on. That's the difference between grading an 8-bit JPEG and grading the full sensor capture.

How it works

rawpy reads the raw sensor data, demosaics it (turning the Bayer/X-Trans pattern into full RGB), applies your chosen white balance and highlight recovery, and decodes into a linear image. A two-stage cache keeps the expensive rawpy decode (~11 s on a 51 MP GFX 50S, per the source) separate from the cheap post-processing, so changing just the camera look or exposure replays in a couple of seconds instead of re-decoding. The optional half_size toggle skips a demosaic step per axis for roughly 10× faster iteration.

The inputs that matter

• raw_file (STRING) - absolute path to the file, picked via the Browse button.
• demosaic - Auto uses DHT for Bayer and LibRaw's 3-pass Markesteijn for Fuji X-Trans. Auto is genuinely the right answer; step down to PPG or Linear only for previews.
• white_balance - As shot (what your viewfinder showed), Auto (gray-world estimate), or fixed Daylight (~5500K).
• highlight_mode - Rebuild reconstructs partial-channel highlight data (Adobe's default); Clip discards, Ignore leaves, Blend sits between.
• output_mode - sRGB display is gamma-encoded and drop-in compatible with every Darkroom node (this is the default you want). Linear scene is pure scene-referred linear for ACES pipelines.
• baseline_exposure (−4 to +8 stops) - in Linear scene mode you'll need this: rawpy's linear decode lands midtones near 0.05, while Camera Raw-calibrated LUTs expect ~0.18. The tooltip's guidance is solid: +2 EV for a daylight scene, more for darker scenes.
• camera_brand / camera_look - the Adobe .dcp profile system. Adobe Standard is the neutral per-body calibration and the default. Pick a brand to unlock that brand's looks. Adobe ships Camera Looks for Canon/Nikon/Sony/Panasonic if Camera Raw or Lightroom is on the same machine; for Fuji film sims you'll need a third-party pack (Stuart Sowerby's or abpy's builders - the README walks through both). If the selected look isn't installed for the detected body, the node silently falls back to Adobe Standard and logs a warning.

Outputs are image and metadata (a RAW_METADATA sidecar for the Split node below).

Install & notes

Part of ComfyUI-Darkroom - Manager, search "Darkroom", or git clone https://github.com/jeremieLouvaert/ComfyUI-Darkroom + pip install -r requirements.txt + restart. Note this one pulls rawpy and exifread from requirements.txt, heavier than the pack's numpy/scipy core; if rawpy fails to build on your Python version, updating it (or the pack) usually fixes it. Use half_size while dialing in your grade, then flip it off for the final render - the author's own workflow.