Run a compatible GGUF language model directly inside ComfyUI. Use OmegaWPN for everyday text generation or turn your ideas into structured image and video prompts with optional H3-style formatting.
Features

• Fully local text generation: No cloud API or subscription required. Models run on your computer through llama-cpp-python.

• Optional H3-style layout: Organize output into visual descriptions, soundscape, and background music.

• Subject and reference tags: Work with text references such as <subject 1>, <picture 1>, <video 1>, and <audio 1>.

• Optional shots: Enable numbered shots and timing instructions, or disable them for continuous prose.

• Flexible directions: Combine subject definitions, shot directions, and general instructions.

• Larger text boxes: Separate areas for instructions, your prompt, and reference or shot directions.

• Generation controls: Adjust temperature, seed, context size, output length, GPU offloading, and more.

• Readable diagnostics: View detected reference requirements without generation stopping over reference-count mismatches.

Getting started

1. Extract the ComfyUI-Demon-GGUF folder into ComfyUI/custom_nodes/.

2. Follow the included README to set up a compatible llama-cpp-python backend.

3. Place your GGUF language models in ComfyUI/models/llm_gguf/, or enter a full model path.

4. Restart ComfyUI, refresh your browser, and search for OmegaWPN.

Before downloading
Models are not included. GGUF compatibility depends on the model architecture and your installed backend.
OmegaWPN generates text prompts. Reference tags describe the media you intend to use; the node does not inspect pictures, watch videos, listen to audio, or generate those media itself.
Output quality depends on your chosen language model. Review generated prompts for invented details, incorrect references, and timing.
An independent community tool—not an official MiniMax integration.
Built for my own workflow, shared for anyone who finds it useful. Enjoy! 🙂

GGUF SFW/NSFW prompting models can be found here: https://huggingface.co/bartowski/Nem...GGUF/tree/main

Source: https://www.360haven.com/forums/downloads.php?do=file&id=1192


Output test:

integrated_multimodal_description: [Shot 1] In the scorching heat of an expansive desert, a lone figure clad in traditional Arabic protective attire trudges across the shifting sands. His movements are naturalistic and lifelike, sweat glistening on his brow as he struggles against the relentless sun.

[Shot 2] At 0:05.364, the camera whips around to reveal a surreal sight: a pirate ship with towering black sails soaring through the sky overhead, its timbers creaking and rigging snapping in the wind. The man stops in his tracks, shielding his eyes from the glare as he gazes upward in disbelief.

[Shot 3] At 2:04.791, the pirate ship's course veers directly towards him, its cannons blazing with a deafening roar and belching smoke. He curses loudly in Arabic, gesticulating furiously at the airborne vessel as he dashes for cover.

N/A

overall_soundscape: The relentless desert wind whips sand against the man's face as he walks, punctuated by his labored breathing and the crunch of shifting sands beneath his feet. The sudden appearance of the pirate ship brings with it the cacophony of creaking timbers, snapping rigging, and the thunderous roar of its cannons.

N/A

non_diegetic_music: N/A


How to use the three text boxes

Top box — Instructions
Tell the model how to write: style, rules, and things to avoid.
Example: “Write a realistic cinematic prompt. Preserve my references. Do not add dialogue, singing, or extra characters.”

Second box — Your prompt
Describe what you want to happen.
Example: “A woman walks through a rainy city at night. The camera follows her as she stops beneath a streetlight.”

Last box — References and shot directions (optional)
Define your subjects, assign media references, or provide specific shot actions. You can combine these or leave the box empty.
Example:

<subject 1> is the woman from <picture 1>.
<video 1> provides the walking motion.
<audio 1> is the sole soundtrack.

[Shot 1] Follow <subject 1> walking through the rain.
[Shot 2] Move closer as she stops beneath the streetlight.

Enable Shots for numbered shots. Disable it for a continuous description, and leave shot directions out of the last box. Subject and reference definitions can still be used. 🙂