Nodes/LCS/LCS Step Observer
ComfyUI Node

LCS Step Observer

LCS Step Observer

By facok·Created 6 months ago·Updated 2 months ago· 160
LCS Step Observer
  • model
  • lcs_data
  • model

This is the pack's microscope. LCS Step Observer doesn't change your image at all - it clones your model, rides along inside sampling, and saves a color-preview PNG for every denoising step to ComfyUI's temp directory. The output is the same MODEL you put in; generation comes out byte-identical.

Why would you want that? Two reasons.

Tuning LCS windows. LCS Color Intervene's default start/end steps (8–10) come from the paper, but your model may behave differently. Watching per-step previews shows you exactly when color "locks in," so you can place your intervention window where it actually does something instead of guessing.

Diagnosing drift. If your sky keeps coming out purple, or LCS Color Anchor is fighting some invisible force, the step sequence shows you where it goes wrong - and it's the difference between fixing a bug and chasing a ghost.

How it works

It installs a post-CFG hook that, at each sampling step, takes the denoised prediction for the first batch item and converts it to an LCS color preview. The clever bit: no full VAE decode. It reads the color coordinates straight out of latent space, which is why it can keep up at every step without wrecking your sampling time. Files land as lcs_step_NNN_sX.XXX.png in ComfyUI's temp directory (the one from folder_paths.get_temp_directory()), one per step, sigma baked into the filename. Then the hook returns the denoised tensor untouched.

Inputs are just model and lcs_data (from LCS Load Data). Output: model. That's the whole interface - no knobs, no optional inputs, no settings to forget.

Finding the files: in a portable build it's under your ComfyUI directory; otherwise wherever your Python environment's temp points. Easiest way to locate them is to search your ComfyUI install for anything starting with lcs_step.

Installing it

Same pack as the rest of LCS - one clone, all nodes:

cd ComfyUI/custom_nodes
git clone https://github.com/facok/ComfyUI-LCS.git
pip install einops safetensors

then restart. ComfyUI Manager works too - search "LCS". Dependencies are just einops and safetensors, both normally already present. (If the nodes don't show up, update ComfyUI; this pack targets the newer V3 extension API.)

Honest take

You will not need this node most of the time. It's a debugging utility for the couple of hours you spend dialing in an LCS workflow - pick your step window, confirm the color is landing, then delete it from the graph. But for that specific job it's exactly the tool you want, and because it's a pure pass-through you can leave it in the graph while you experiment without paying any image-quality cost. Just remember the previews only cover the first item in a batch, so multi-color batch work needs a look at each run, not one observer.

CategoryLCS/observe

Inputs (2)

NameTypeDefaultDescription
modelMODEL
lcs_dataLCS_DATA

Outputs (1)

NameTypeDescription
modelMODEL