tools
A personal collection of ComfyUI workflows, installers and benchmark notes for running MiniMax H3 video generation and Qwen Image 2.1 image generation locally on a four-GPU RTX 3090 rig, plus a standalone hand-drawn animation skill.
This repository collects tools the author built while running generative models on personal hardware, cleaned up for others to use, with each section self-contained and independently installed. A MiniMax H3 section provides thirteen ComfyUI workflows for open-weight video-with-audio generation, one set for any single 24 GB card and another that spreads a multi-GPU box across all its cards, along with the custom nodes and a patch needed for its turbo LoRA to load.
A Qwen Image 2.1 section covers image generation, editing and transparent PNG output on a single RTX 3090, with native ComfyUI workflows, an isolated installer, SHA-256 model verification, a script that submits a generation and checks the output, and a lower-VRAM GGUF variant for smaller GPUs. A separate skills section includes a hand-drawn canvas animation skill for making short animated films in JavaScript and Canvas 2D, with editable examples, an MP4 renderer and browser regression checks, requiring no external rendering engine or video-generation model.
All measurements in the repository's documentation come from real runs on the author's own four-card RTX 3090 rig.