Power Prompt (rgthree)

Power Prompt is a prompt box that also does the boring plumbing around a prompt: it gives you dropdowns to insert embeddings and LoRAs by name, it can encode the text to conditioning right there, and it can even parse <lora:...> tags out of your prompt and load those LoRAs for you - no separate loader nodes. If you liked writing <lora:name:0.8> inline the way you did in Automatic1111, this is the node that brings that back to ComfyUI.

Why you'd reach for it

Stock ComfyUI splits prompting into pieces: a text box here, a CLIP Text Encode there, LoRA loaders somewhere else, and you have to remember the exact filename of every embedding and LoRA because there's no menu. Power Prompt folds all of that into one node. You get a big multiline text field, a dropdown to pick embeddings so you never fat-finger a filename, a dropdown to insert LoRA tags, and even saved prompt snippets you can drop in. It's the "I just want to write my prompt and go" node.

The neat part is how flexible its wiring is. You can use it three ways depending on what you connect:

• Connect nothing and it acts like a fancy String node - it outputs the raw TEXT for some other node to handle.
• Connect a CLIP and it encodes the text itself, handing you CONDITIONING directly.
• Connect a MODEL and it parses any <lora:...> tags in the text and loads them automatically - no separate LoRA loader in the graph at all.

Inputs and outputs

The one required field:

• prompt - your multiline text. Write it like a normal prompt; inline <lora:...> tags are honored if a model is connected.

The optional inputs are what unlock the extra modes: opt_model (connect to enable inline LoRA loading), opt_clip (connect to encode to conditioning), and three dropdown helpers - insert_lora, insert_embedding, and insert_saved - which don't change the output type, they just paste the right text into your prompt for you.

Outputs cover every mode: CONDITIONING (populated when a CLIP is connected), pass-through MODEL and CLIP (so you can keep the wire going after LoRA loading), and TEXT (the raw processed string). Wire CONDITIONING into your KSampler's positive input; pass MODEL onward to the sampler if you used inline LoRAs.

Installing it

Power Prompt is part of rgthree-comfy. Install through ComfyUI Manager by searching rgthree's ComfyUI Nodes, or clone the repo:

cd ComfyUI/custom_nodes
git clone https://github.com/rgthree/rgthree-comfy.git

Restart ComfyUI. No models to download and no heavy dependencies - the pack is pure graph and UI code.

Common issues

The classic gotcha: inline <lora:...> tags do nothing if you haven't connected a model. Without opt_model wired in, the node treats the text as text and passes your LoRA tags through as literal characters. Connect a MODEL and pass the node's MODEL output onward, and the loading actually happens.

Second, the usual LoRA confusion that isn't Power Prompt's fault at all: a LoRA that seems inert is usually missing its trigger word, not failing to load. Load it, but also type its keyword into the prompt - many LoRAs are dormant until you do. (rgthree's Power Lora Loader has a right-click info dialog that surfaces trigger words, which is the community's favorite way to track them down.)

Note also that negative prompts don't support LoRAs - for a clean negative, use Power Prompt - Simple, the LoRA-free sibling built for exactly that.

And the pack-wide caveat: if the node's dropdowns or text area render wrong, check whether ComfyUI's Nodes 2.0 frontend is on. The Vue rewrite broke several rgthree nodes, and the standing workaround is to leave Nodes 2.0 disabled.