Local $0 Video Head Swap: MiniMax H3 Ref2VA (INT8) + LoRA in ComfyUI
Melvin Vivas · X video post · 2026-10-08 · 0:17 · 1,276 views · Open on X
Topics: AI Dev Tools & Productivity, Fine-tuning & Model Customization · Level: intermediate
Summary
This is a short progress update on an experiment: swapping faces/heads in video on a local machine at no cost. Melvin Vivas runs the MiniMax H3 Ref2VA model, quantized to INT8, inside ComfyUI, adds a head-swap LoRA, and fits it on one RTX 3090 with VRAM peaking at 23.8 GB. He says the results aren't perfect yet. The post gives no workflow, settings or code. The video has no useful speech.
Key points
- Goal: run video face/head swapping fully on your own machine, with no API or cloud costs ($0).
- Model: MiniMax H3 Ref2VA, a reference-to-video model, quantized to INT8 so it uses less memory.
- Runtime: ComfyUI, a node-based workflow tool for generative models.
- A head-swap LoRA is added on top of the base model to focus it on swapping heads.
- Hardware: one RTX 3090 (24 GB), with VRAM peaking at 23.8 GB, which is nearly all of the card's memory.
- Quality isn't perfect yet; it's a work in progress.
- The creator says he's working on it together with Grok (xAI's assistant), using an AI chatbot as a helper for the experiment.
- The post shares no workflow file, LoRA name or settings.
Resources mentioned
- MiniMax H3 · tool · huggingface.co · free
A MiniMax reference-to-video generation model, run here as an INT8 quantized version for local video head swapping.
Also in: Let an Agent Run Overnight Local Video Generation with ComfyUI (Melvin Vivas on X · notes), Making a 40-Second Anime Episode Locally with ComfyUI on an RTX 3090 (Melvin Vivas on X · notes), Free local anime video generation with Minimax H3 on an RTX 3090 (Melvin Vivas on X · notes), A day of work done from a phone with Grok Bot agents (Melvin Vivas on X · notes) and 4 more - ComfyUI · tool · x.com · free
The official X account for ComfyUI, the node-based tool for generative AI workflows, which announces things like ComfyUI Agent.
Also in: Grok as "Director": Making a Local Anime Clip with ComfyUI on an RTX 3090 (Melvin Vivas on X · notes), FLUX 3 Image Klein: Open-Weights Image Model Now Supported in ComfyUI (Melvin Vivas on X · notes), ComfyUI Templates: Chain Ready-Made Workflows Instead of Building From Scratch (Melvin Vivas on X · notes), Comfy Agent: An AI Agent That Builds ComfyUI Workflows in Comfy Cloud (Melvin Vivas on X · notes) and 32 more - Grok · tool · x.com · free
xAI's LLM chatbot; the creator runs a custom 'ML Grok Bot' on it to help debug and improve the pipeline.
Also in: Grok Bot Turns an X Post into a Shareable Dashboard (Melvin Vivas on X · notes), Demo: Grok Turns Your X Post Data into a Dashboard (Melvin Vivas on X · notes), Using a Custom Grok Bot Persona to Learn Arduino (Melvin Vivas on X · notes), Using a Grok Custom Bot Template as an Arduino Tutor (Melvin Vivas on X · notes) and 32 more - NVIDIA GeForce RTX 3090 · tool · nvidia.com · paid
A consumer GPU with 24 GB of VRAM, used as the local hardware for image and video generation.
Also in: Running an Overnight 100-Iteration Agent Loop with a Local Qwen Model (Melvin Vivas on X · notes), Grok as "Director": Making a Local Anime Clip with ComfyUI on an RTX 3090 (Melvin Vivas on X · notes), Making a 40-Second Anime Episode Locally with ComfyUI on an RTX 3090 (Melvin Vivas on X · notes), Portable Computer Now Runs AI Agents and Models Locally on NVIDIA RTX PCs (Melvin Vivas on X · notes) and 6 more
Try this
- Build a fully local, $0 video head/face-swap pipeline in ComfyUI with an INT8 quantized video model plus a head-swap LoRA, and track peak VRAM on a 24 GB GPU.
More in AI Dev Tools & Productivity
- Why Model Speed Matters in Coding Agents: Codex vs ChatGPT Plans
- Claude in Google Docs, Sheets and Slides: An Alternative to Gemini
- GitHub Copilot Adds Intelligent Local Model Routing (Project HydraFusion)
- Pairing Grok Bot with Cursor Cloud Agents for productivity
- Running Grok Bot with ComfyUI locally from an iPhone
- Microsoft MAI-Code-1.1-Flash: On-Device Coding Model in GitHub Copilot