realrebelai/rebelspromptenhancer

Rebels Prompt Enhancer is a set of add-on nodes for ComfyUI, a popular desktop application for running AI image and video generation locally. The nodes take a rough text prompt you type in and automatically expand and improve it before passing it to the image generator, aiming to produce better results without you needing to learn the specific phrasing each image model prefers. The enhancement is done by a small language model called Qwen3.5-4B that runs entirely on your own machine. Nothing is sent to external servers. The model is provided in a format called GGUF, which is designed to run with low graphics memory usage. Two size options are available: a smaller version around 1.5 to 1.8 GB for users with limited GPU memory, and a larger 6 GB version that produces better output. The main node works through three dropdown menus. The first sets the purpose: whether you are generating an image, editing an existing image, or generating video. The second sets the output format, matching the syntax expected by whichever image model you are feeding the prompt into. Different image models use different prompt conventions, and the node has presets for many common ones including Flux, SDXL, Stable Diffusion 1.5, and several video models. The third sets an aesthetic style, chosen from 22 options such as cinematic film, watercolor, cyberpunk, or anime. A second node lets you load any GGUF model file you provide, with full control over sampling settings. A third node takes an image as input and generates a prompt describing it, useful for extracting a starting point from a reference image. A fourth node acts as a checkpoint in your workflow that displays the current prompt on screen and can pause the pipeline until you confirm the result. Installation involves cloning the repository into the ComfyUI custom_nodes folder and installing dependencies. The required model files are downloaded separately from a public model hosting site and placed inside the node folder. The README includes links and size information for each supported model.