🖼️ → 📝 PromptBuilder by Nexus

Drop an image. Get a prompt. That simple.

A standalone Gradio app that turns any reference image into a detailed, generation-ready prompt — powered by local Ollama vision models running entirely on your machine. No cloud, no API keys, no subscriptions.

16 style presets — from Krea2 photorealistic to MidJourney, Booru tags, cinematic film stills, and even video prompts with camera movement control for LTX and Wan workflows.

Smart length control — from a quick one-liner to a full detailed paragraph. Built-in token budgeting keeps the output exactly the length you asked for.

JoyCaption-style extra options — 17 toggles for lighting, camera angle, composition, depth of field, SFW/NSFW classification, and more. Same controls you love, better package.

LoRA-ready — trigger word field auto-prepends to every prompt. Character name support built in.

VRAM-friendly — one-click model unload frees your GPU for generation. Or set it to auto-unload after every prompt so you can bounce between prompting and generating without thinking about it.

Works with Qwen3-VL, JoyCaption, and any other Ollama vision model you throw at it. Double-click start.bat and you're running — venv creates itself on first launch, no setup headaches.

Part of the Nexus AI toolset.