AI-HUB — Local Electron app for ComfyUI, with VLM chat, voice, and an autonomous video director

A free, open-source desktop app that wraps ComfyUI in a friendly UI and adds the things that make AI generation actually fun to use day-to-day: a smart chat with image-gen tool use, voice in/out, a preset-based model+LoRA system, and an autonomous AI Director that strings together long-form videos clip by clip.

Built around Wan 2.2 SVI Pro for seamless 1+ minute videos with end-frame continuation, but works with any SDXL / Qwen / Wan checkpoint you already have.

What's inside

• Studio — manual T2I and I2V with preset-driven model + LoRA stacks, HireFix upscaling, body-fix pass

• AI Director — autonomous pipeline: Planner → Smart LoRA Scan → T2I → Body Fix → HireFix → I2V → Extend ×N → Planner Review with retries. Premiere-style timeline of generated clips.

• Chat — VLM conversation that can call <generate> to produce images inline, with Whisper STT and Kokoro/Piper TTS for voice in/out. TTS fires in parallel with image gen so it feels instant.

• Presets — sidebar-based preset management with descriptions, LoRA pools, and smart preset selection by VLM

• Gallery — browse generated, imported, and Director project content

• First-run wizard — one-click install of the VLM runtime, voice server, Whisper, and Kokoro. Detects your GPU and applies a recommended VLM profile.

Stack

Electron 28 + React 18 + Vite 5 + Tailwind 3 + Zustand on the front. Python FastAPI for the VLM (llama-cpp-python with CUDA wheel) and Voice (Whisper.cpp + Piper/Kokoro) backends. 226 Vitest unit tests across the pure-logic modules.

Requirements

• Windows / Linux (macOS untested)

• Node.js 18+

• Python 3.11

• ComfyUI running locally

• NVIDIA GPU with CUDA, 8 GB+ VRAM recommended for VLM inference

Get it

GitHub: https://github.com/alhajrimh883-ai/AI-HUB/tree/main

The first-run wizard handles the rest.

License

MIT — use it, fork it, ship it.