SigmaSync LoRA accepts the same SIGMAS tensor used by the sampler and assigns a LoRA strength to every model-call sigma. It supports generated curves and exact per-sigma strengths without merging the LoRA into the base model weights.

Installation

cd ComfyUI/custom_nodes
git clone https://github.com/capitan01R/ComfyUI-SigmaSync-LoRA.git

Restart ComfyUI after installing or updating.

No additional Python packages are required beyond the dependencies already provided by ComfyUI.

Included Nodes

Node	Outputs	Purpose
SigmaSync LoRA	`MODEL`	Loads a model LoRA and changes its strength against the scheduler's exact sigma trajectory.

Usage

Use a scheduler or custom-sigmas node that outputs SIGMAS, such as Basic Scheduler.

Model Sampling -> Basic Scheduler or Custom SIGMAS -> sampler (sigmas)
                                      \-------------> SigmaSync LoRA (sigmas)
Model Sampling -------------------------------------> SigmaSync LoRA (model) -> guider

Connect the same SIGMAS output directly to both SigmaSync LoRA and the sampler. SigmaSync reads the schedule but does not modify it.

For a scheduler tensor containing nine values, the sampler has eight model-call sigmas. Explicit mode therefore requires eight strength values. The terminal endpoint is not a model call.

sigmas:             1.0000, 0.9567, 0.9045, 0.8403, 0.7595, 0.6546, 0.5128, 0.3109, 0.0000
explicit_strengths: 0.75,   0.75,   0.75,   0.75,   1.20,   1.40,   1.20,   1.30

Controls

Parameter	Default	Meaning
`lora_name`		LoRA selected from the ComfyUI model LoRA directory.
`start_strength`	`1.0`	Strength at the first model-call sigma for generated curves.
`end_strength`	`0.1`	Strength at the last model-call sigma for generated curves.
`curve`	`linear`	Selects `linear`, `cosine`, `smoothstep`, `power`, or `explicit`.
`curve_power`	`2.0`	Used by `power`. Values above `1` hold the starting strength longer.
`explicit_strengths`		Used by `explicit`. Enter one comma- or space-separated strength per model-call sigma.

In explicit mode, explicit_strengths overrides start_strength, end_strength, and curve_power.

At exact scheduler sigmas, the corresponding configured strength is used. If a sampler evaluates between two supplied sigmas, the strength is linearly interpolated between those anchors in sigma space.