# MiniMax H3 LongVideos

## Summary

Same-day MiniMax H3 long-video workflow with synced audio and shot chaining.

## Model Family

- `minimax`

## Status

- `imported-only`

## Source

- Model card: <https://huggingface.co/Smite79/MiniMax-H3-Longvideos>
- Workflow file: <https://huggingface.co/Smite79/MiniMax-H3-Longvideos/resolve/main/workflows/H3%20LongVideos%20Demo%20Workflow.json?download=true>
- Original publisher: `Smite79`
- Date imported: `2026-09-07`

## Developer notes

- The workflow is built around the `H3 LongVideos` node pack.
- It turns a paragraph-by-paragraph prompt into chained shots and supports `plan_only` for a non-rendering split check.
- The source README expects ComfyUI `0.31+` with native MiniMax-H3 support.

## Our notes

- This is the most immediately useful MiniMax H3 import today because it packages the long-form, audio-synced path instead of a bare checkpoint.
- The default settings are already aimed at practical use: `8` steps, `res_multistep`, `shift_video/shift_audio` at `12/3`, and shot chaining.
- Still heavy for an RTX 3070 with `8 GB` VRAM, but this is the right kind of heavy if you want a real H3 workflow.

## Required custom nodes

- `Smite79/MiniMax-H3-LongVideos`
- current ComfyUI MiniMax-H3 support
- `pysssss` for the text output panels

## Required models

- `MiniMax-H3` diffusion model
- `MiniMax-H3` text encoder
- `MiniMax-H3` video VAE
- `MiniMax-H3` audio VAE
- optional `MiniMax-H3` PDD Acc LoRA for the low-step path

## Notes

- Best first try if you want a reusable local MiniMax H3 workflow rather than another checkpoint.