alesha-pro/tools

Tools, ComfyUI workflows and benchmark configs from a 4x RTX 3090 local-inference rig

What it solves

This repository provides a collection of specialized tools for running AI models on local hardware, specifically focusing on optimizing video generation and creating agent-based animations.

How it works

The project is organized into independent sections, each with its own installation process:

  • minimax-h3: Implements ComfyUI workflows for the MiniMax H3 model, which generates synchronized video and audio in a single pass. It includes optimizations for both single 24 GB GPUs and multi-GPU setups, achieving significant speedups on four-card configurations.
  • skills: Contains standalone agent skills, such as a JavaScript and Canvas 2D tool for creating hand-drawn canvas animations (pencil, ink, risograph, etc.) without requiring Blender or video-generation models.

Who it’s for

Users running local AI inference on their own hardware (particularly those with NVIDIA RTX 3090s or similar multi-GPU setups) who want to generate high-quality video and audio or implement specific animation skills for AI agents.

Highlights

  • Multi-GPU Optimization: Specifically tuned for 4x RTX 3090 setups, reducing sampling time for 15-second shots from 19 minutes to under 7 minutes.
  • Sychronized Generation: Utilizes MiniMax H3 to produce frames and audio simultaneously.
  • Agent Skills: Includes a specialized animation tool for expressive strokes and authored poses using Canvas 2D.
  • Ready-to-Use: Ships with custom nodes, necessary patches for turbo LoRA, and simplified installers.

Related

  • Project
    Alos21750/UAV-Downloader

    UAV Downloader is a Windows‑focused video downloader for JableTV, MissAV, SupJav and Hanime1 that can automatically generate Japanese, English and Traditional Chinese subtitles using local speech‑recognition (ReazonSpeech K2 v2 or Whisper) and translation models (FuguMT, OPUS‑MT). It offers three operation modes – an interactive GUI (UAV Browser), an unattended scheduler (UAV Watcher), and a headless Docker/CLI version – with proxy support, parallel downloads, and optional LLM‑based translation. The project is released under Apache 2.0 and runs offline by default.

  • Project
    MME-Benchmarks/Video-MME-v2

    A robust benchmark for evaluating video understanding in MLLMs, using a three-level progressive difficulty scale and grouped non-linear scoring to measure true temporal reasoning.

  • Project
    AlexGladkov/claude-in-mobile

    `mcp-devices` (aka `claude‑in‑mobile`) is a modular Node/Rust server that lets Claude‑style AI agents control Android, iOS, web, desktop, Aurora, and HarmonyOS devices. Install the base (`npm i -g mcp-devices`), add platform plugins (e.g., `@mcp-devices/plugin-android`), enable them, and point an MCP‑compatible client at the server. The tool provides actions like screenshots, taps, UI inspection, live debugging, and testing, enabling AI‑driven device automation and embodied‑agent research.

  • Project
    11273/mooc-work-answer

    A cross‑platform Python tool that automates learning‑progress tracking, discussion posting and AI‑assisted homework hints on Chinese MOOC services (智慧职教, AI 优课, 资源库). It talks to the official APIs and optionally calls DeepSeek to generate reference answers (60‑100 % relevance). The app can be run as a pre‑built executable or from source, and includes safety delays to mimic human usage.