Load Settings CSV

Load Settings CSV is the "where does this take place" node of the ComfyUI-CSV-Loader pack. Its settings.csv ships with around a hundred world-and-scenario presets - Ancient Curse, AI Uprising, Age of Exploration, Biopunk - and turns them into one dropdown. Pick a setting, get a positive prompt out the other side, wire it to your CLIPTextEncode, and suddenly your "city street at night" scene has a thesis. It's the cheapest way to give a batch of generations a consistent mood without retyping the same three sentences every time.

Don't overthink what it does, because there's genuinely nothing to overthink. The name is not a lie about calling a settings API or needing config - it reads a plain CSV and hands you strings. That's the whole node, and it's better for being that simple.

How it works

On load, the node opens custom_nodes/ComfyUI-CSV-Loader/CSV/settings.csv, skips the header row, and parses each line into three columns: name, a positive prompt, and a negative prompt. The parser is CSV-aware, so quoted commas inside a prompt won't split your line - a row like Steampunk,"gears, brass, steam, ", stays intact. Each name becomes an entry in the node's dropdown, and picking one just looks up the matching positive and negative strings and hands them to you. No model files, no API calls, no Python dependencies beyond what ComfyUI already ships. Install it and it works; that's the whole maintenance story.

One thing to know: the dropdown is built when the node loads, not live. Edit the CSV and you need to hit Refresh in ComfyUI before the new entries show up - the author's own error messages literally tell you to do this.

The inputs and outputs that matter

• settings - the only input. A dropdown listing every name in settings.csv. That's it.
• positive prompt (STRING) - the preset's positive text. Wire this into the positive CLIPTextEncode in your workflow.
• negative prompt (STRING) - usually empty in the shipped file, because most rows were added with no negative. You'll often be feeding an empty string here.

The beginner trap: a STRING output won't go straight into a KSampler. It has to pass through a CLIPTextEncode node (that's the node that turns text into conditioning the model understands), and the text encoder differs per model - tags for SDXL-lineage checkpoints, sentences for Flux and the LLM-encoded models. The preloaded settings are written in comma-tag style, so they behave best on SDXL-family checkpoints.

Installing it

The pack ships as a zip, which is how the author originally distributed it - unzip into ComfyUI/custom_nodes/. Or use ComfyUI Manager (search "ComfyUI-CSV-Loader"), or clone it:

cd ComfyUI/custom_nodes
git clone https://github.com/PCMonsterx/ComfyUI-CSV-Loader

Then restart ComfyUI. No pip install, no model downloads, no extra weights - this is one of the rare custom nodes with zero dependencies, which also means it can't break your environment the way heavier packs can.

Common issues

• The dropdown only shows "Error loading settings.csv, check the console". The file isn't where the code expects it. The path is hardcoded to the folder name ComfyUI-CSV-Loader - if your clone or unzip got a different name, the mapping breaks. Keep the folder exactly as it ships.
• One bad line kills the whole file. The parser builds a dict in one pass, so a row with two columns instead of three (or a stray unclosed quote) fails the entire load, not just that row. Keep three columns per line: name,positive,negative.
• You edited the CSV and the dropdown didn't change. Refresh ComfyUI; the choices are cached from when the node loaded.

Worth knowing before you build a workflow around it: the pack's README says the author has stopped maintaining it and points to Dr.Lata nodes and the forked projects for the same functionality. It's not broken - this node is so simple that "unmaintained" is nearly a compliment - but don't expect new settings.csv entries from upstream. Add your own; that's the point of it.