Latent Get Channel Means

A measurement node, not a manipulation one: feed it a latent, and it computes the average value across each of that latent's channels, returning them as a list. It's the diagnostic half of RES4LYF's channel-wise toolkit - where LatentBatch_channels and LatentBatch_channels_16 let you push channel values around, this node tells you what those values currently are, on average, across the whole latent.

The practical uses are all downstream of "I need a number before I can decide what to do." Check whether a latent has drifted toward one color axis before deciding how much to correct it with LatentBatch_channels. Compare the channel means of two different latents (a guide versus a generation, or two frames of a video) to see how far apart they are before reaching for something like Latent Match Channelwise. Or just use it as a sanity check when a result looks off and you want a number rather than a guess about what's happening in latent space.

Input and output

• latent (LATENT) - required, the latent to measure.
• Output: channel_means (SIGMAS) - one value per channel, in the pack's generic float-list type. SIGMAS is normally a noise-schedule type in ComfyUI, but RES4LYF reuses it here as a plain array - that's why the output type looks unusual for what is really just a list of averages.

Nothing to configure - it's a pure read, no side effects on the latent itself.

Installing it

Bundled with the full RES4LYF pack:

• ComfyUI Manager - search "RES4LYF," install, restart.
• Manual - cd ComfyUI/custom_nodes && git clone https://github.com/ClownsharkBatwing/RES4LYF, then cd RES4LYF && pip install -r requirements.txt, restart.

No extra downloads for this node.

Common issues

Not sure what to do with the output. This node only measures - it doesn't feed directly into a visual result on its own. Pair it with something that can display SIGMAS values, or feed it into another RES4LYF node (like the schedule inputs on LatentBatch_channels) that actually consumes a per-channel array.

Numbers don't obviously map to anything visible. Latent channel values don't correspond one-to-one with pixel-space brightness or color in a simple way - a mean near zero doesn't necessarily mean "neutral" in any visually intuitive sense. Treat this as a relative comparison tool (before versus after, this latent versus that one) rather than a value with an absolute, human-readable meaning on its own.

Output length doesn't match what you expected. The number of values in channel_means matches however many channels your latent actually has - 4 for SD1.5/SDXL-style latents, 16 for Flux, Z-Image, and similar newer architectures. If you're expecting one count and getting another, check which VAE produced the latent in the first place.