RebelsPromptEnhancer Offers Local Prompt Boost For Private Workflows

    
        By vramkickedin    
     | 
    
            May 31, 2026 at 10:45 pm        
    
     | 
    
        2 min read

RebelsPromptEnhancer is a new ComfyUI node pack that rewrites short text ideas into detailed prompts using a lightweight 4-billion-parameter language model, running entirely on your own hardware. It works offline with no API keys or cloud calls, and it aggressively frees VRAM the moment the enhancement finishes so your diffusion model gets the GPU back. The pack includes a curated enhancer, an experimental custom-model loader, an image-to-prompt vision node, and a workflow-locking gate.

Built by independent developer RealRebelAI and shared openly on GitHub, the project gives privacy-conscious users a completely local alternative to online prompt services. The creator paired a layered style system with multiple model-format presets so a single 4B model can produce output tuned for Flux, SDXL, Pony, video models, and more. The four nodes are kept ultra-lightweight to run well on consumer GPUs without interfering with the main image or video generation.

Local prompt control with layered styles and caching

Key Features

100% local, no data leaves your machine.
Aggressive VRAM cleanup after every call.
Layered style: purpose, model format, aesthetic.
Lock mode caches prompt, skips model reload.
Custom GGUF node for any local text model.
Image-to-prompt node with vision models.
Workflow gate displays the locked prompt text.
22 aesthetic styles from photorealism to anime.

The tool fits freelancers, small studios, and serious hobbyists who need private prompt enhancement without sending client ideas to a remote API. The careful VRAM management makes it practical for 8GB cards, and the lock-and-iterate pattern helps teams freeze a preferred prompt while tweaking other generation settings. Anyone who wants repeatable workflows and zero ongoing costs will find the offline caching and gate system immediately useful.

Developer notes and experimental edges

The developer openly marks version 2 as a work in progress with more features coming, and several nodes carry experimental warnings. The custom GGUF enhancer can hand back bad results depending on the model and sampler settings, so users are advised to test temperature and top-p until the output stabilizes. The image-to-prompt node still needs refinement, but chaining its caption with the dedicated text enhancer often produces better results than letting a vision model do both jobs.