🗂️Dict → Sigmas

The advanced one, for people who bus their samplers

Dict → Sigmas is a narrow little node: it pulls a SIGMAS object back out of a dict. If you're not building a "bus" workflow that carries whole sampler setups around, you may never touch it. If you are, it's exactly the tool for the job.

Quick background on SIGMAS, because it's easy to have seen the term without knowing what it is. In the newer ComfyUI node API, the noise schedule - the ramp of noise levels a sampler steps down through during generation - is a first-class object produced by scheduler nodes and fed into the sampler. The Sigmas you pick effectively set the pacing of the denoise. When you're assembling a full sampling setup into a dict (guider, sampler, sigmas, maybe noise), you need a typed extract node to get the sigmas back out at the other end without losing the type.

What it does

Under the hood it's dict[key] with a type check bolted on. Everything that applies to the other extract nodes applies here:

• dict - the DICT to pull from.
• key - the exact name the sigmas were stored under.
• cleanup_key - on by default; strips spaces and extra newlines from the key you type, so "sigmas " doesn't accidentally fail to match "sigmas".

The value output is a SIGMAS wire, which plugs straight into any sampler node that accepts a sigmas input in the new API. You don't get a number or a tensor you can read - you get the schedule object, ready to be consumed.

Installing it

Part of the 🗂️ Dict Tools pack. Easiest via ComfyUI Manager - search Dict Tools (Lex-DRL/ComfyUI-DictTools):

cd ComfyUI/custom_nodes
git clone https://github.com/Lex-DRL/ComfyUI-DictTools

Restart after cloning. The pack's only dependency is frozendict, and there are no model downloads. Two version caveats: the pack requires ComfyUI >= 0.18.0, and SIGMAS as a first-class type only exists in the newer node schema - on older ComfyUI neither the node nor the type will be there, so update ComfyUI if the nodes don't register.

Where people get burned

The type check is strict. If the value stored under that key isn't actually a Sigmas object, execution fails rather than silently coercing it - so make sure the matching Add node was a DictAddSigmas, not a generic Any add that happened to hold something sigmas-like. And as with every extract node here, a key that doesn't exist raises a No such key in Dict error, not a gentle empty result. Keys are case-sensitive; get them exactly right the first time and save yourself the traceback.

It's a small node for a specific habit. But if your workflow busses advanced sampling configs around, this is the one that makes the bus work without a type conversion fight at the end.