Sign In

Guide | DaSiWa MiniMax H3 Director and Workflow

17

Guide | DaSiWa MiniMax H3 Director and Workflow

Features

I'm very excited to write this! - The H3 Director and Workflow got huge updates 💥
Timestamp: 2026\09 - Workflow version v2.4
Repo: https://github.com/darksidewalker/ComfyUI-DaSiWa-Nodes

Workflow: https://civitai.red/models/2831978/dasiwa-minimax-h3-workflows-or-t2va-or-fl2va-or-ref2va

I want to highlight the following -

The ♾️ Continuity and ✨ Prompt Forge was a collaboration with TrashAI and thearas3248 from my Discord community!

This enables the Director to seamless expand on videos endlessly and make prompting absolute effortless with connected LLM's.

The H3 Director now holds a crazy amount of features:

  • 🎬 MiniMax H3 modes: T2VA, I2VA, L2VA, FL2VA, REF2VA and single-frame Image Inpaint in one workflow

  • 🖼️ Reference Director: image, video and audio lanes with drag-reorder, per-media prompts and first-frame video thumbnails

  • 🎵 Audio references: standalone tracks, external video soundtracks and embedded-video V / A / V+A selection

  • ⏱️ Media trimming: draggable crop markers, playable crop preview and validated reference-duration limits

  • 📋 Fast media input: file-manager drag-and-drop, lane-selected Ctrl+V and paste-replace on a selected tile

  • 📸 RefMods: saved image, video and audio references, including v5 bundles, with strength and editable descriptions

  • ✍️ Prompt editor: one free-text prompt per mode with optional structure, shot insertion and reference-label prefill

  • ✨ Prompt Forge: draft and review prompts with a local LLM, Ollama or an OpenAI-compatible server; vision models can inspect reference pictures

  • ♾️ Continuity: optionally extend a completed H3 video/audio take with a separate next-action prompt

  • 📐 Smart resolution: automatic or custom aspect, resolution and megapixel presets with Torch Resize input scaling

  • 💾 Reference packs: save and load media, prompts and RefMod selections with append or overwrite

  • 🧩 Native H3 generation: Director and Guide route to ComfyUI’s built-in H3 nodes with selected-model lazy loading

Guide

All settings and in-depth explanations are on the GitHub!

This is a global overview over the quick accessible options and handling the director.

Basic Usage

Bildschirmfoto_20260926_110125-1.png

1# Setup your sampling settings and select the model files you want to use

2# Add any LoRA's you might want

3# Set your SEED

Bildschirmfoto_20260926_110325.png

4# Set the duration (ref_image_size and frame_rate should not be changed if you do not know what you do.)

5# Pick the mode you want MiniMax H3 to run

6# Pick your resolution, aspect and RefMod's - Auto will pick the MiniMax model defaults and automatic aspects of the first media

7# Add any media you want to use or reference, dependent on the mode you use

Write a prompt, use ♾️ Continuity and/or ✨ Prompt Forge and hit run


Advanced Usage

Use ✨ Prompt Forge

Bildschirmfoto_20260926_112315.png

Add your LLM - Server

Bildschirmfoto_20260926_112440.png

8# Write a brief explanation what you want

9# Generate - let the magic happen - and Apply

Use ♾️ Continuity

Bildschirmfoto_20260926_112755-2.png

10# Activates ♾️ Continuity for starting after the video generation
or

11# Let you pick a already produced video and start continuity from this
or

12# Picks an already made checkpoint

13# Write the prompt or use ✨ Prompt Forge to tell MiniMax how to continue the video

  • If nothing is inserted a basic follow prompt is used

Choose one and expand your video journey into eternity 💫


Closing Words

The goal of my Node Pack is to bring you as much as possible features as easy to use.

I always try to make it with as less dependencies as possible and self-containing as much as I can. This is all my hobby and many things would not be possible without my awesome followers and many dedicated friends and members of my community supporting me!

Thank you for everyone who supported me! 💌

dxCuide [TOUS], WirW

CWW23r

17