Prompt Mixer dAIly v2 (Dynamic CSV)

The v1 prompt mixer in this pack is a solid CSV wildcard node, but it has one structural quirk: the attributes it knows about are hardcoded to the example columns, so a CSV with different headers leaves you stuck inside its little box. Prompt Mixer dAIly v2 (Dynamic CSV) is the author's fix, and it's a good one. Your CSV headers become the placeholders - rename a column and the node follows - so any CSV you already have works instead of one that matches the template.

What changed, and why it's worth it

Everything in v2 flows from one idea: read the headers, build the node around them. The big wins:

• Headers as placeholders. If your CSV has a column called Setting, you write {Setting} in the template. No fixed inventory of attribute names to stay inside.
• Dynamic inputs and outputs. The node grows attr_001…attr_064 ports to match your columns (or however many it needs), and an ATTR_* output per column so you can wire a specific picked value elsewhere in the graph.
• Empty columns are allowed. A column can stay empty and its value gets supplied from upstream nodes instead - handy when one attribute shouldn't come from the file.
• File picker + refresh. You can pick a CSV from a browser-style dialog, which copies it into the pack's local csv/ folder, and a Refresh button reloads the current file without re-copying. The template autocompletes {...} from the loaded headers, and the node keeps its size when you reload.
• Toggle the noise. show_attribute_outputs hides the attribute ports when you only care about the prompt, which keeps the graph from turning into a cable spaghetti monster.

Inputs that matter

• csv_path (STRING) - the CSV to load, absolute or relative to the ComfyUI root. v2 keeps the same path semantics as v1.
• template (STRING, multiline) - write it with {HeaderName} placeholders matching your CSV columns.
• flat_mode (BOOLEAN) - ON merges everything to a single line; OFF preserves your paragraphs and line breaks.
• show_attribute_outputs (BOOLEAN) - toggle the per-column output ports on or off.
• seed (INT) - pick the same seed, get the same combination. This is the whole reproducibility story, so set it deliberately if you plan to re-run or compare batches.
• schema_json (STRING) - the internal schema the dynamic ports are built from; you'll normally leave it as [] and let the node populate it from the CSV.

Outputs are PROMPT (STRING, always the first one - wire this to your text encoder), USED_JSON (STRING, the picked values plus seed for your metadata), and the dynamic ATTR_* outputs. That USED_JSON is the feature that keeps you honest in a batch: attach it to the image metadata and "which prompt made this one?" stops being a mystery.

A quick note on v1 vs v2

If you're starting fresh, start with v2 - it does everything v1 does plus the dynamic headers. The author kept v1 around for existing workflows, not because it's better. If you're using v1 only because a shared workflow references it, swap and don't look back.

Installation

Same pack, same install: ComfyUI Manager, search "dAIly Prompt & Token Utils" (or just "dAIly"). Or manually:

cd ComfyUI/custom_nodes
git clone https://github.com/DailyMok/ComfyUI-dAIly.git

Restart ComfyUI. No model files or heavy dependencies.

Gotchas

Two things trip people up. If you change a CSV's headers after the node loaded, the ports and placeholders can go stale - hit Refresh (the template default literally tells you to "click Refresh CSV after changing headers") before you blame the output. And keep the encoding happy: UTF-8-SIG is preferred, so save from Excel as "CSV UTF-8" or you'll be debugging mojibake in your options instead of generating.