Extensions/ComfyUI-MiniMaxH3-FlowDirector
ComfyUI Extension

ComfyUI-MiniMaxH3-FlowDirector

Director flow for ComfyUI, custom node that enalbes generation of long videos without OOM by using the last frame of each generation as initial for a new one.

By AlonAshken·Created 7 days ago·Updated 7 days ago· 19
AlonAshken/ComfyUI-MiniMaxH3-FlowDirector
Nodes4
On cloudLocal install
CategoryMiniMax H3
Stars19
Updated7 days ago

Nodes (4)

MiniMax H3 Enhance Prompt

Let a local vision model write your MiniMax H3 prompt — and describe the exact images it will see

MiniMax H3
MiniMax H3 Flow Director
MiniMax H3
MiniMax H3 Preview Override

Actually watch MiniMax H3 denoise, instead of a single frozen frame

MiniMax H3
MiniMax H3 Save Last Frame

Keep the last frame of every MiniMax H3 shot — the one you feed into the next

MiniMax H3
Readme

ComfyUI-MiniMaxH3-FlowDirector

License: MIT ComfyUI

ComfyUI-MiniMaxH3-FlowDirector is an advanced WYSIWYG visual timeline director for MiniMax Hailuo H3. It enables the generation of infinitely long, continuous videos with synchronized audio by breaking long timelines into discrete sequential flows, passing the decoded last frame as the first frame of subsequent blocks, and transitioning towards target images—all with a low, constant VRAM footprint.

✨ Key Features

  1. Discrete Sequential Flows (No OOM / No Crashes):

    • Instead of trying to sample a giant 1000-frame video all at once, each timeline segment is sampled and decoded in isolated blocks.
    • Low and constant memory footprint across any length (from 10 seconds to 10+ minutes).

  2. Automated Last-to-First Frame Chaining:

    • The final decoded frame of Block $i$ is automatically passed as the opening keyframe (first_frame) to Block $i+1$, guaranteeing continuous motion flow.

  3. Target Transition Frames (last_frame):

    • Dragging an image onto Block $N > 0$ automatically treats that image as the destination transition frame. MiniMax H3 smoothly morphs and transitions from the previous block towards your target image!

  4. Seamless Seam Stitching:

    • Redundant duplicate boundary frames between consecutive blocks are dropped automatically to ensure smooth playback without seam stutter.

  5. Integrated Audio & Motion Tracks:

    • Full timeline audio waveforms with automatic trimming, stereo mixing, and native MiniMax audio decoding.

*You can chain workflows by having just text, image to text and then just more text prompts, or controlling director parts with more images that will be used as "last-frame" for previous chunks to lead to, which I think is the best way to do it.

Optimal video length is 5-10 seconds for each chunk (and you can chain however many you want, leading up to multiple-minute seamless videos).

📦 Installation

Method 1: Via Git Clone

Open your terminal in ComfyUI/custom_nodes/ and run:

cd custom_nodes
git clone https://github.com/AlonAshken/ComfyUI-MiniMaxH3-FlowDirector.git

Method 2: Via ComfyUI Manager

Search for MiniMax H3 Flow Director in ComfyUI Manager and click Install.

🚀 Recommended Workflow Integration

[FL2VA Model]        ───> [model       ]
[REF2VA Model]       ───> [model_ref2va]
[MiniMax Text CLIP]  ───> [clip        ]  [MiniMax H3 Flow Director]
[MiniMax Video VAE]  ───> [vae         ]  ─────────────────────────> images ───> [VHS Video Combine]
[MiniMax Audio VAE]  ───> [audio_vae   ]  ─────────────────────────> audio  ───> [                 ]
[KSamplerSelect]     ───> [sampler     ]  ─────────────────────────> fps    ───> [                 ]
[BasicScheduler]     ───> [sigmas      ]
[RandomNoise]        ───> [noise       ]

An example workflow is included in example_workflows/MINIMAX_H3_FLOW_DIRECTOR_ULTRA_TURBO.json.

📄 License

MIT License - see LICENSE for details.