Prompt Manager Text

When you want the prompt, not the embedding

Prompt Manager Text is the sibling of Prompt Manager with the encoder removed. No clip input, no CONDITIONING output - you type a prompt, and you get the final text back as a plain STRING. It still does the part that actually makes this pack worth installing: every prompt is auto-saved to the same SQLite database with categories, tags, and image links, and the admin dashboard treats it identically.

Reach for it when the prompt's destination isn't a sampler. Maybe you're sending text to a style-applying node, a text processor, an LLM that rewrites prompts, or any node that only accepts STRING. Or you're building a workflow where the prompt is assembled dynamically and you want to see - and store - exactly what got built, without the overhead of encoding it. If you do need conditioning in the same graph, note the full PromptManager node also exposes a STRING output alongside CONDITIONING, so the encoding version covers this case too.

How it works

The mechanics mirror the encoding node. On execution it joins prepend_text + text + append_text into one string, optionally injects LoRA trigger words if you've enabled the LoRA Manager integration, saves the original text to the database (SHA256 dedup - identical text updates the same row), and returns the combined string. It's also set to always re-execute rather than cache, so every run re-processes and re-saves. That's the behavior you want from a logging node.

Inputs: text is required (multiline). Optional: category, tags, prepend_text, append_text, and a search_text widget - again, actual searching belongs to PromptSearchList and the dashboard, not this widget. One output: STRING.

The prepend/append pair is the sleeper feature. Put the style block you keep reusing in append_text, keep the thing you're iterating on in text, and you can change the mood of a whole set of prompts from one node without touching every prompt. It's also how this node composes with others: PromptManagerText → text processor → CLIPTextEncode is a perfectly normal chain.

Installing it

Same pack, same steps - one install covers all three nodes:

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

or search "Prompt Manager" in ComfyUI Manager, then restart. Dependencies are watchdog and Pillow; nothing model-sized downloads for this node (the AutoTag models are separate and only pulled on demand).

Pitfalls

The classic mistake is wiring this where a CONDITIONING is expected - you'll get a type error, because it's text-only. If you need both outputs, use the encoding node. An empty text isn't saved to the DB, so a blank run just passes the prepend/append through. And a database write failure is logged, never fatal: the string still comes out the other end.