📥 NL Read

ComfyUI's built-in Load Image is fine for a single PNG. It's useless the moment your shot is a 150-frame EXR sequence, a video file with a specific in/out range, or a plate that needs to land in a specific colorspace before anything else touches it. That's the gap NL Read exists to fill. It's the node from this pack that studio folks actually build pipelines around, and once you've used it, going back to Load Image feels like going back to MS Paint.

NL Read is the "I" in the pack's color-managed IO story: read a source, normalize it into the working space (ACEScg by default), and hand the graph both the image and the metadata that says what it is. It also leans on the NL Workflow context panel - set your project, shot, resolution, and frame range there, and NL Read can pull frame ranges straight from that context instead of you typing numbers everywhere.

How it works

Point source at a path or pattern and the node figures out whether it's a still, a sequence, or a video. The controls that matter most:

• range_mode - context uses the workflow's start/end frame, custom lets you type your own, off ignores ranges.
• every_nth - sample every Nth frame, the cheap way to preview a sequence at half or quarter rate.
• preframes / postframes - duplicate the first or last frame, handy when you need a video model to hold on a frame.
• reverse_sequence - read frames backwards.
• force_resize - resize to context resolution or a custom size, with fit/fill/stretch strategies.
• input_colorspace - auto-seeded from the source you load, and it's the actual interpretation used. Leave it alone unless you know better.
• output_colorspace - where the emitted tensor lives, ACEScg by default.
• preview_colorspace - display-only. It affects what you see in the preview, never the tensor.

Outputs worth knowing: image, mask and inverted_mask (from the source's alpha or your mask mode), frame_count, resolved_path (the actual file it resolved - great for debugging pattern mismatches), and image_info_json, which carries the colorspace and source metadata onward. Wire that string into NL Grade, NL Write, or any of the color nodes and the pipeline stays coherent.

Installing and setting up

Same pack install as everything here: ComfyUI Manager, search "ComfyUI-NL_Nodes", or clone manually:

cd ComfyUI/custom_nodes
git clone https://github.com/NOLABEL-VFX/ComfyUI-NL_Nodes

Restart, then open the top-bar NL Workflow panel and set at least a project path before you lean on context-driven reads. Video reading needs ffmpeg on the system; the color-managed path wants PyOpenColorIO and OpenEXR from requirements-color.txt.

Where people get burned

The big one is path policy. NL Read ships in guarded mode: it blocks obvious system locations like /etc, /proc, /sys, /dev, and the Windows system roots, which is a sane default for a node that can read arbitrary files. But it also blocks things that aren't malicious - your NAS mount or a studio share. If reads fail on a path you're sure exists, set NL_READ_EXTRA_ROOTS to whitelist it. The NL_READ_ALLOW_ANY=1 env var exists to restore old unrestricted behavior, but treat that as a debugging escape hatch, not a lifestyle. And if you're coming from the wider ComfyUI ecosystem, remember this pack isn't a big Reddit name - it's a studio tool, so the polish is real but the community walkthroughs you're used to just don't exist yet.