LTX-2 Control Preprocess (aux)

Control-style workflows almost always start with a preprocessor - the thing that turns a plain image into a grayscale, an edge map, or a thresholded silhouette that the model can steer by. IAMCCS LTX-2 Control Preprocess is a lightweight version of exactly that, bundled as the README describes it: a "lightweight grayscale/threshold/edges helper for control-style workflows." No model downloads, no heavy dependency tree, just six transforms on an IMAGE.

The README calls it out as part of the LTX-2 utility set, but honestly it's generic - if you're doing any aux/control pipeline (or even just debugging a mask-ish preprocessing chain), it works the same.

How it works

mode selects the transform:

• none - passthrough, nothing happens.
• grayscale - desaturate.
• invert - flip the values.
• threshold - hard cutoff at threshold (0–1, default 0.5): below becomes black, above becomes white.
• edges_sobel (default) - Sobel edge detection, no extra dependencies.
• edges_canny (opencv_if_available) - Canny edges. The name is honest: it uses OpenCV if it's available, and falls back gracefully if not. Canny's two knobs are canny_low (default 100) and canny_high (default 200).

There's a genuinely thoughtful detail in the source: it defensively normalizes the input. ComfyUI IMAGE tensors are normally float in [0,1], but some custom nodes hand over [0,255] or even [-1,1]. A naive preprocessor clamps that and crushes the signal to all-white or all-black; this node detects the range and re-normalizes before transforming. That's the kind of robustness that stops an entire category of "why is my control image blank" bugs.

Outputs: image (the processed frames) and report (a text summary of what mode and parameters ran).

What it's for

If you're building LTX-2 aux-control workflows, feed it a reference frame and wire the edge output where your control conditioning expects its input image. If you're not doing control at all, grayscale and threshold still earn their keep as quick prep stages before a mask or a comparison. It's the sort of node that's unremarkable until you realize you've been reaching for a huge ControlNet-preprocessor pack to do what one combo widget does.

Install

It ships in IAMCCS-nodes:

cd ComfyUI/custom_nodes
git clone https://github.com/IAMCCS/IAMCCS-nodes.git

or ComfyUI Manager → search IAMCCS, restart. The only optional extra is OpenCV if you want real Canny instead of the Sobel fallback.

The honest note

This is a convenience preprocessor, not a replacement for the dedicated ControlNet preprocessor packs - it has no pose/seg/depth estimators, and it won't autodownload anything. If your workflow only needs edges and thresholding, it's more than enough and a fraction of the install weight.