Load Positive CSV

Load Positive CSV is the odd one in the ComfyUI-CSV-Loader pack: its positive.csv ships with exactly one preset, called Realistic, whose positive prompt is 8K, Realistic, Hyperrealism, Depth of Field,. That's it. One entry, a handful of quality tags. If you're here expecting a big menu of positive-prompt recipes, the honest review is that this node is more of a template than a library.

But don't dismiss it, because the thinness is the feature. This is the node the pack gives you to make your own. A single named entry that returns a positive string is a clean way to bolt a "quality baseline" onto a workflow - the tags that should be in every generation go in the file once, and any workflow using this node inherits them. Change the baseline in one place and every downstream generation updates. For a "masterpiece, best quality"-style habit that you never want to retype, that's genuinely handy.

How it works

Same engine as every node in this pack. It reads custom_nodes/ComfyUI-CSV-Loader/CSV/positive.csv, skips the header row, and parses each line into name, positive prompt, negative prompt. The header in the shipped file is name,prompt,negative_prompt, so the positive output is what the column header calls "prompt". The dropdown lists the names; picking Realistic returns its tag string. The negative output is empty.

Because the whole value is the file, the practical move is to edit it before you do anything else. Open it in a text editor or spreadsheet app, keep the header, and add rows in the same three-column shape:

name,prompt,negative_prompt
Realistic,"8K, Realistic, Hyperrealism, Depth of Field, ",
My Baseline,"sharp focus, high detail, ",

Keep quoted commas inside a field (the parser is CSV-aware) and every row at exactly three columns, then hit Refresh in ComfyUI so the dropdown rebuilds.

The inputs and outputs that matter

• positive - the dropdown. Names from positive.csv.
• positive prompt (STRING) - into your positive CLIPTextEncode.
• negative prompt (STRING) - empty unless you add one.

And the usual wiring rule: STRING output feeds a CLIPTextEncode, not the sampler directly. The shipped tags are comma-style, which suits SDXL-lineage checkpoints; on the newer LLM-encoded models they'll be read as flat words and carry less weight.

Installing and caveats

Zero dependencies, like the rest of the pack: unzip into ComfyUI/custom_nodes/, install via ComfyUI Manager (search "ComfyUI-CSV-Loader"), or git clone https://github.com/PCMonsterx/ComfyUI-CSV-Loader, then restart. Keep the folder name exactly ComfyUI-CSV-Loader - the loader's path is hardcoded to it, and a renamed folder gives you the telltale "Error loading positive.csv, check the console" dropdown.

The pack is unmaintained (the author points to Dr.Lata nodes and forks), but that matters least here: this node's entire job is to read a file you control. The one shipped preset is a starting point, not the product.