Multi Note (Combine All)

If you've ever stared at a 200-word prompt jammed into one CLIP Text Encode box and lost track of which clause does what, this node is the fix. Multi Note (Combine All) turns your prompt into a stack of little labeled blocks - one for subject, one for pose, one for environment, one for style - and stitches them all back into a single string when it's time to generate. It's a prompt editor that lives inside the graph, not a text file on your desktop.

It ships in ComfyUI-M-Nodes by oMo.Studio, a small utility pack whose whole pitch is "manage huge prompts without losing your mind." The name is the promise: combined_text out, ready to feed a CLIP Text Encode.

How it works

The node starts with one block: a toggle (enable_1), a title (title_1), and a multiline text area (text_1). The trick is what happens when you type - the moment you put text in the last block, the node's front-end JS spawns a fresh triplet (toggle, title, text) below it, up to as many as you need. Empty blocks don't pile up; they only appear once you've actually written something in the one before.

On the backend it's dead simple. When you hit Queue, it walks every enable_N/text_N pair, keeps the blocks whose toggle is on and whose text isn't blank, and joins them with ", " into one string. That string is the combined_text output.

The inputs that matter

• enable_N - the toggle per block. Off means that block is skipped entirely. Great for A/B-ing a clause without deleting it.
• title_N - pure organization. It's a label so you can tell "Pose" from "Environment" at a glance; it's not part of the output.
• text_N - the actual prompt content.

One output: combined_text (STRING). Wire it wherever a prompt string goes.

The master-prompt pattern

The README's own suggested workflow is worth copying:

1. Add a Multi Note, make blocks for Subject, Pose, Environment, Style.
2. Right-click a Random Prompt Selector (also in this pack) and choose Convert text to input.
3. Connect combined_text to the Selector's text input.
4. Feed the Selector's output to your CLIP Text Encode.

Now your big prompt is both organized and randomized - you can put {red|blue|green} variations inside any block and let the selector shake it up. If you're on SDXL or Flux, a structured prompt like this is also just easier to debug than one wall of text.

Installing it

ComfyUI Manager: search "ComfyUI-M-Nodes" and install. Or do it by hand:

cd ComfyUI/custom_nodes
git clone https://github.com/oMoStudio/ComfyUI-M-Nodes

Then restart ComfyUI. There's no requirements.txt and no model to download - this is pure JS plus a small Python node, so it's one of the friendlier installs in the ecosystem. MIT licensed.

Where people get burned

• Commas are the separator, so mind the join. Blocks are combined with ", ". That's fine for prompts, but a block that ends mid-word or a tag list where you don't want a comma between two sections will read awkwardly.
• Blank and disabled blocks vanish silently. If every block is toggled off, you get an empty string out, and your CLIP Text Encode will happily encode nothing.
• The dynamic fields are the whole point, and they only exist in the UI. This is why the node reads the graph's widget state rather than a fixed schema - it's the same trick a few other "grows as you type" nodes use, and it means the fields do survive save/reload (that's the pack's headline feature), but it's worth knowing the blocks are front-end generated. On the Nodes 2.0 frontend rewrite, dynamic-widget packs like this are the ones that occasionally render oddly - if a loaded workflow looks off, re-add the node rather than fighting it.

It won't generate anything on its own. But if your prompt workflow has become unreadable, this is the cheapest reorganization you'll find.