⭐ Star Conditioning Loader

You've hit this exact wall: a workflow that takes thirty seconds of setup time on a good GPU, and a third of it is the text encoder spinning up and re-encoding a prompt you've run a hundred times. Star Conditioning Loader is the other half of the StarNodes save/load pair - the half that reads a previously-saved conditioning file back off disk and hands it to a sampler. No CLIP node in the graph, no encoder loading, no waiting.

It's in ⭐StarNodes/Conditioning and pairs with its sibling Star Conditioning Saver, which writes the files it reads.

How it works

The single input is a dropdown, conditioning_file, which lists every .pt file in ComfyUI/output/conditionings/ sorted newest-first. Pick one and the node loads it and outputs a CONDITIONING - the exact prompt embedding that was saved, ready to wire into a sampler's positive (or negative) input.

If the folder is empty (or the pack is fresh and you've never saved anything), the dropdown shows No conditioning files found and the node returns an empty conditioning rather than erroring - which is polite, but means your sampler gets nothing to condition on. So the practical rule is: it's a loader, not a generator. It only works if something saved first.

Why it's useful

Because conditioning is a prompt embedding, not a token - it's already been through the CLIP/text-encoder and come out as a tensor. Loading it skips the most expensive single step in many modern workflows. That's why this node (and its saver) exist in a pack aimed at people who run the same base prompts over and over: for a fixed character prompt or a signature style prompt you re-use constantly, you encode once and stop paying for it.

Installing it

Part of the StarNodes pack - ComfyUI Manager, search Starnodes, install, restart:

cd ComfyUI/custom_nodes
git clone https://github.com/Starnodes2024/ComfyUI_StarNodes
cd ComfyUI_StarNodes
pip install -r requirements.txt

Search the canvas for star - it's under ⭐StarNodes/Conditioning.

The gotchas

The big one, which the loader can't protect you from: a saved conditioning is locked to the model family it was encoded with. Load an SDXL-era embedding into a Flux sampler and you'll get garbage or an error, and the node won't warn you - the file looks fine, it's just incompatible. Keep your saved conditionings organized by model. Second, the dropdown lists everything in the folder newest-first, so once you've saved a few variants you need to read the timestamped names carefully to grab the right one - the author's timestamp-on-save design trades overwrite-safety for a slightly cluttered list. Prune the folder now and then and this is a genuinely nice node.