Channel selector for 16 channels

Latent spaces come in different widths, and if your model's noise predictions run 16 channels deep, the 4-toggle selector from this pack isn't going to cut it. Channel selector for 16 channels is the pack's answer for that middle tier: sixteen channel_0 through channel_15 booleans, all on by default, one CHANS output. It's a plumbing node with a single purpose - describe which channels a downstream pre-CFG node should operate on.

The only consumer it's built for is the channels_selection input on Pre CFG automatic scale (and the other CHANS inputs this pack exposes). Feed the output in there and the auto-scale node will restrict its per-channel normalization to the channels you've left enabled. Turn one off and that channel is simply skipped by the automatic logic - which is exactly the move when you've identified that a single channel is the source of a color cast or a recurring artifact. The README's framing for these nodes is that order and chaining are yours to tune; the channel selectors are the fine-grained dials inside that tuning.

All the selectors in the family are generated from one base class, so the behavior is identical across the 4/8/16/32/64/128 versions - the only difference is how many channels they describe. Sixteen is a nice practical middle: if your model uses a 16-channel latent, this is the right size, and it's still a manageable grid of toggles on screen (unlike the 128 version, which is a wall of switches you'll only touch for genuinely wide latents).

A note on matching: the selector's job is to describe a channel count, and it doesn't validate against your model. If the consuming node's predictions have fewer channels than your selector describes, the list just won't line up - so pick the selector matching your model's actual channel count. For most SDXL and SD 1.5 work that means the 4-channel version; reach for 16 when you know your latent is wider.

Install with the rest of the pack:

cd ComfyUI/custom_nodes
git clone https://github.com/Extraltodeus/pre_cfg_comfy_nodes_for_ComfyUI

or via ComfyUI Manager searching pre_cfg_comfy_nodes_for_ComfyUI, then restart ComfyUI. No model downloads, no extra Python packages.

Don't expect fireworks from this node by itself - it's a quiet dependency that earns its place when wired into the automatic scale. It won't be the star of your workflow, but it's the knob that makes the star behave.