LATENT To Any 🔽

A LATENT is the compressed in-progress state of a diffusion run - the thing between the sampler and the VAE. It's not an image yet, and you'd usually never route it somewhere generic. But ComfyUI's strict typing occasionally puts you in a spot where a latent needs to go through an * (any) socket. This node takes the LATENT, passes it through untouched, and declares the output as any. No VAE decode, no noise changes, no pixel output.

It's one of the ComfyUI Telegram Suite pack's generated converters, alongside its AnyToLATENT reverse and a full row of XToAny nodes for every common type.

Why it exists

Same story as the rest of the converter category: the suite's trigger inputs and outputs accept any type, and before version 1.0.4 you had to cast your signal to ANY before feeding a trigger and convert it back after. That's why the category is so full - the README all but apologizes for it. Since 1.0.4 the conversion is no longer required for triggers; the changelog keeps the converters for backward compatibility.

The real remaining use is a wildcard-input node that won't take a typed LATENT output. When the connection refuses and you're sure the target genuinely accepts any object, this is the cast.

How it works

A pass-through, pure and simple:

def convert(self, LATENT):
    return (LATENT,)

The latent dict (with its samples tensor and any noise_mask) is returned as-is. The LATENT input is a forceInput socket, and the output is any of type *.

Inputs and outputs

• LATENT (required) - the latent to pass through.
• any (output, type *) - the same latent, typed as any.

Installation

Pack-wide:

cd ComfyUI/custom_nodes
git clone https://github.com/SwissCore92/comfyui-telegram-suite.git
cd comfyui-telegram-suite
pip install -r requirements.txt

or install "ComfyUI Telegram Suite" via ComfyUI Manager and restart.

Notes

Don't confuse this with anything that turns a latent into a viewable image - that's the VAE Decode job. This node does zero heavy lifting and that's the point. And if it's sitting in a trigger chain from an old workflow, it's harmless but optional; fresh graphs feed triggers directly.