Home Creators Posts Import Register Favorites Logout

Downloads

Content

Video : https://youtu.be/RIFTcIZqaS4

Hey everyone.

New video is up: MiniMax H3 Might Be The Best Local AI Video Now! Multimodal Reference2Video Walkthrough.

MiniMax finally dropped H3 open weights on Hugging Face and ModelScope, and ComfyUI already has merged / quantized files ready for consumer GPUs. This is not just another text-to-video model. It is an omni-modal setup — text, images, video, and audio — and in the video I walk through what actually runs locally right now (H3 base / 768p path), plus the honest limits vs Seedance-style fight-scene character replace.

Free workflows attached for the three modes I demoed:

1. video_minimax_h3_t2v.json — text-to-video (FL2VA)

2. video_minimax_h3_i2v.json — image-to-video / first-frame style (FL2VA)

3. video_minimax_h3_r2v.json — multimodal reference-to-video (Ref2VA)

These use the pruned Int8 Convrot weights + Qwen3-VL 32B NVFP4 text encoder, with separate video and audio VAEs. Good starting point if you want to try H3 without hunting every file yourself.

Paid post has the fuller breakdown — what each graph is doing, model file list, frame-length formula notes, tips, and common issues from the recording session.

— Benji

MODELS REFERENCED

Video : https://youtu.be/RIFTcIZqaS4

- MiniMax H3 (official weights): https://huggingface.co/MiniMaxAI/MiniMax-H3

- ComfyUI merged / quantized files: https://huggingface.co/Comfy-Org/MiniMax-H3

- Diffusion variants: FL2VA (text-to-video / image-to-video), Ref2VA (reference-to-video + audio)

- Text encoder: Qwen3-VL 32B (BF16 / Int8 Convrot / NVFP4)

- VAEs: MiniMax H3 video VAE + audio VAE (separate files)

- Note: Distillation LoRA not available yet at time of recording; default sampling ~20 steps

Files

MiniMax H3 Might Be The Best Local AI Video Now! Multimodal Reference2Video Walkthrough

This video walks through MiniMax H3 after the open-weight release on Hugging Face and ModelScope, with a full ComfyUI setup. We cover the omni-modal generative system, FL2VA vs Ref2VA weights, Qwen3-VL 32B text encoder options (BF16, Int8 Convrot, NVFP4), separate video and audio VAEs, and why the marketed 2K path is really H3 base at 768p plus H3 Regenerate 2K on a second pass. You get text-to-video demos, reference-to-video with character sheets plus audio, prompt-guide tips with AI agents, and honest limits on fight-scene character replace versus Seedance. This is for ComfyUI users and local AI video creators who want MiniMax H3 on a single consumer GPU without a cluster. If you already run quantized diffusion models, care about VRAM choices (Int8 Convrot vs FP8 Scaled vs BF16), or you want reference-to-video with up to 9 images, 3 videos, and audio in one workflow, this walkthrough is aimed at you. Also useful if you care about private or on-prem generation for data sovereignty. Open weights for MiniMax H3 change what you can run locally for multimodal video: text, images, video, and audio in one model, with native stereo audio and long prompts under 7,000 characters. ComfyUI already has merged and quantized files, so you can start on hardware like a 3060 (slow) or a workstation GPU (about 2 minutes per 5-second clip in this demo). Beyond entertainment, MiniMax is also pointing H3 at world-model style robotics prediction. For ads, music-video style clips, and multi-character talking scenes with audio sync, this is one of the strongest open options right now, with clear gaps still left for hard V2V fight replace. MODELS REFERENCED Blog post: https://www.patreon.com/aifuturetech/posts/165674768?utm_source=youtube&utm_medium=video&utm_campaign=20260804 - MiniMax H3 (official weights): https://huggingface.co/MiniMaxAI/MiniMax-H3 - ComfyUI merged / quantized files: https://huggingface.co/Comfy-Org/MiniMax-H3 Timeline : 00:00 – Introduction to MiniMax H3 and its open-weights release. 00:41 – Exploring the "Omnimodal" generative system: Text, Image, Video, and Audio. 01:15 – Understanding the two-stage generation process (768p base to 2K output). 02:08 – Summary of specifications and system requirements. 02:43 – ComfyUI setup: Selecting the right model files (INT8 vs. FP8). 04:19 – Diving into the Text Encoder (Qwen 3VL 32B) and VAEs. 05:09 – Prompting guide: How to use AI agents to enhance your prompts. 06:29 – Real-world applications: MiniMax H3 as a "World Model" for robotics. 07:12 – ComfyUI Workflow: Text-to-Video generation walkthrough. 08:18 – Visual Showcase: Coffee close-ups and character skin details. 09:34 – Breaking down the subgraph: Frame formulas and sampling steps. 11:13 – Motion Showcase: Animal documentaries and high-action BMX clips. 12:30 – Advanced Tutorial: Reference-to-Video and multimodal tasks. 13:36 – Testing character consistency and audio-influenced motion. 14:44 – Case Study: Character replacement in fight scenes (Pros & Cons). 16:17 – Final thoughts: Privacy, data sovereignty, and the future of local AI video. Local Workstation GPU : https://amzn.to/3XfXsAO -------------------------------------------------------------------------------------------------------------------------------- If You Like tutorial like this, You Can Support Our Work In Patreon: https://www.patreon.com/c/aifuturetech #minimaxai #minimaxaivideo #comfyui #aianimation

Comments

Hj

this is amazing i have a question, how big is your character sheet for it to be so crisp? like what pixels are they, very nice video

Pastil

"MiniMaxH3ImageToVideo" nodes?

Pastil

Many manual patches cannot, Thanks ill try later!