Model (Improved)

AugmentWaitModel - "Model (Improved)" in the menu - is the trigger-aware passthrough for the MODEL type in Augment's Flow Control family. The MODEL is your loaded checkpoint or LoRA-stacked base model, and it's the heaviest, most important thing moving through your graph. When you swap checkpoints or stack LoRAs in a multi-stage workflow, ComfyUI's output cache can decide the model "didn't change" and happily run stages on the old weights. This node pins that down.

The stock way to pass a model is a wire from your loader. And it usually works - until the caching short-circuit kicks in. The Wait node adds the two things a wire can't: a trigger token that sequences the sampler stages, and a guarantee (via the pack's executor patch) that it always re-runs instead of being pruned as unchanged.

How it works

The name is a bit of a lie - it doesn't hold the model until a trigger arrives. From the source, execute passes the model straight through and emits a trigger token. The real mechanism is the pack's executor override: on import, Augment patches ComfyUI's execution internals so any node that takes or emits a TRIGGER type always re-runs and is never skipped by the output cache. Chain trigger outputs into the next stage's trigger input and your model path runs in the order you designed, every run.

That "always re-runs" bit matters more for MODEL than any other type, because it's the one value where "I loaded a different checkpoint" has to actually reach the sampler.

The inputs and outputs

• value (MODEL, required) - the model to carry. Comes in from your checkpoint loader or model-stacking nodes; there's no widget here.
• trigger (optional) - the sequencing input from a previous stage.
• Outputs value (MODEL), json (AUGMENT_JSON), trigger (TRIGGER) - the model passes through untouched, the trigger continues the chain, and json carries a structured dump (type and value as a string) for Augment's JSON Extract / JSON Viewer.

Nothing to configure - pure ordering and re-execution, which for the model is exactly what you want.

How to install

Augment isn't on the ComfyUI registry yet, so it's a manual clone - and mind the folder name:

cd ComfyUI/custom_nodes
git clone https://github.com/augment-lib/augment-ComfyUI augment
cd augment
pip install -r requirements.txt

The clone target must be augment. The pack imports everything as custom_nodes.augment.py.*, so the default folder name breaks loading. Restart ComfyUI afterward; nodes appear under Augment.

Dependencies: opencv-python, numpy, torch, Pillow, requests - already installed from running ComfyUI. No model files, no API key (only the pack's vector nodes use the paid Augment API).

Common issues

• Checkpoint swap "not taking effect" - cache behavior. Thread the model through this node and sequence the triggers so the sampler re-runs on the new weights.
• Node missing after install - folder name. Rename to augment, restart.
• Everything re-executes now - by design. Trigger nodes always re-run; keep Wait nodes where deterministic re-runs matter.