EmberFrame Attach Sigma To Latent

This is the most boring node in the EmberFrame pack, and that's the compliment. EmberFrame Attach Sigma To Latent takes a LATENT and a FLOAT sigma and returns a latent that carries that sigma embedded inside it. That's the whole job.

Why does that exist? Because the PiD workflow has a coordination problem. Nvidia's PiD is a pixel-diffusion decoder - a learned module that replaces (or supersizes) a normal VAE decode, and it wants to know which sigma the latent you're handing it was captured at. In the EmberFrame staged-decode path, that number comes out of the capture node as a separate wire (pid_sigma), and you've got to get it from the sampler all the way over to PiD Prepare without losing track of which latent it belongs to.

This node is the answer: instead of dragging a parallel FLOAT wire across your graph and hoping it stays paired with the right latent, you bake the sigma into the latent dict itself under a pid_sigma key. Any downstream EmberFrame node that cares - the Latent Capture Inspector reads it, the Normalize node reports it - can read it straight off the latent.

When you actually use it

The capture sampler already returns pid_latent with the sigma attached, so you don't need this node in the happy path. You reach for it when you're assembling a latent from elsewhere:

• You took a stock SamplerCustomAdvanced output (which knows nothing about PiD sigmas) and want to feed it to a PiD decode path.
• You're reconnecting a workflow after it got shuffled and a latent lost its sigma on the way through.
• You want to force sigma 0 on a latent - which is exactly right for a final denoised latent going into a final-latent PiD decode.

Set sigma to the value you know the latent was produced at. Zero is the safe default and the right value for a fully denoised output; for intermediate captures you want the sigma that was actually in effect at that step.

Install & wiring

It's part of EmberFrame Nodes, under EmberFrame → PiD:

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

Restart ComfyUI. No models, no dependencies, nothing to download - the whole pack is torch-and-Comfy-only.

The output latent wires straight into PiD Prepare or any other latent consumer. That's it. This node is a glue stick, and glue sticks don't need a troubleshooting section. If your PiD output looks wrong, the problem isn't the sigma was attached - check the Normalize node's constants and the PiD scale rule (scale = 4 for Z-Image/Flux) first.