Prompt Manager

If you've ever rebuilt a prompt by squinting at an old PNG's metadata, this is the node that ends that habit. Prompt Manager (class PromptManagerAdvanced) is the flagship of the ComfyUI-Prompt-Manager pack: a save/load prompt library that also carries your LoRA stacks, trigger words, and a thumbnail, all in one box. The author built it because the other prompt managers he found felt too heavy - it shipped as a single node, and one of the first people to try it said it replaced their old manager outright.

You wire it inline with your normal graph. Pick a saved entry from the category and name dropdowns and it pushes the prompt text and LoRA stacks downstream; pick a blank name, type into text, and hit save when a generation works. It's an output node, so it's also happy sitting at the end of a workflow just to record what you did.

How it works

Prompt Manager keeps two independent LoRA stacks, lora_stack_a and lora_stack_b, plus a prompt string, saved per category. The dual-stack thing exists because Wan-style video workflows genuinely want separate stacks (one for the UNet, one for the text encoder or per-model). The use_lora_input mode tells it how to treat what you connect:

• Prompt LoRAs Only - use the stacks saved with the prompt entry
• Combine LoRAs - merge connected stacks with the saved ones
• Input LoRAs Only - ignore saved stacks, use only what you wire in

use_prompt_input is the equivalent switch for text: flip it to use a connected prompt instead of the stored one. swap_lora_outputs just swaps the A/B outputs - handy when you've built a workflow for one ordering and a saved entry assumes the other. Each input accepts either a LORA_STACK or a MULTI_LORA_STACK, so the multi-slot stackers in this pack slot right in.

The outputs are prompt, lora_stack_a, lora_stack_b, multi_lora_stack (the merged pair as one payload), and recipe_data - that last one lets the prompt feed the pack's Recipe toolset, which is how you move from "this prompt worked" to "this whole pipeline is reproducible."

The inputs that matter

• category / name - the library. Name is the long list of saved prompts; category filters them.
• trigger_words - comma-separated tokens appended to the prompt, useful with character/style LoRAs that need their magic words.
• thumbnail_image - connect an image and it's saved as the thumbnail for this entry, so the picker shows you faces, not text.

Installing

Install the pack once via ComfyUI Manager (search "Prompt Manager", author FranckyB), or:

cd ComfyUI/custom_nodes
git clone https://github.com/FranckyB/ComfyUI-Prompt-Manager.git
cd ComfyUI-Prompt-Manager
pip install -r requirements.txt

Then restart ComfyUI. Prompt Manager itself needs nothing extra - no models, no API keys. The llama.cpp backend only matters for the Prompt Generator nodes in the same pack.

Where people get burned

The most common confusion is the LoRA mode. If you connect stacks and nothing changes, check use_lora_input - the default is Prompt LoRAs Only, which deliberately ignores your connected stacks. Second: saved data lives in your ComfyUI user folder, and the pack auto-rotates backups on startup, so an overwrite is rarely fatal but is still worth confirming - the UI asks before it deletes anything. If you want LoRA previews on hover and trigger words read off the file automatically, install ComfyUI-Lora-Manager alongside it; this node lights up when that pack is present.

Should you use it?

If you're on SDXL/Wan and juggling more than a few reusable prompts, yes - this is the one to reach for. If you never reuse prompts, you don't need a prompt manager at all, and that's fine too. The recipe_data output quietly makes it the front door to the pack's bigger claim: turn any prompt you like into a full renderable recipe.