MiniMax H3 Chain Director V1.0

MiniMax H3 Chain Director V1.0, also known as H3 Director Studio V1, is a custom ComfyUI Director designed to control complete MiniMax H3 audiovisual workflows from one centralized interface.

It combines BASE generation, latent 3D UPSCALE, multi-segment video generation, audiovisual continuity and live previews without removing the external ComfyUI connections required by advanced users.

Download

Download the public workflow JSON using the Download button on this Civitai page.

The custom Director node, source code, installation instructions, dependencies and technical documentation are available in the official GitHub repository:

GitHub:
https://github.com/M1cr0Bytes/MiniMax-H3-Chain-Director-V1

V1.0 Release:
https://github.com/M1cr0Bytes/MiniMax-H3-Chain-Director-V1/releases/tag/V1.0

Why use the Chain Director?

Generate longer videos

Chain mode divides a project into separate segments and executes them sequentially.

Each segment can use its own prompt while still receiving the general project instructions. This makes longer and more structured videos easier to direct and manage than a single oversized generation.

Maintain audiovisual continuity

The Director integrates MiniMax H3 Motion Context to transfer audiovisual context from one segment to the next.

Continuity states can be saved and restored between segments, while the context-trimming system helps remove repeated transition frames before the next segment begins.

Control the workflow from one panel

The central interface provides access to:

• General Prompt

• Independent segment prompts

• Segment creation and selection

• Single and Chain modes

• Video duration

• Aspect ratio

• BASE resolution

• BASE and UPSCALE sampling steps

• Independent samplers and schedulers

• Fixed or randomized seeds

• Preview BASE and Final render modes

• Upscale multiplier

• Independent BASE and UPSCALE models

• Independent BASE and UPSCALE LoRAs

The controls are centralized, but the external ComfyUI connections remain visible and editable.

Configure BASE and UPSCALE independently

The workflow uses a two-stage pipeline:

1. BASE generates the initial MiniMax H3 video.

2. UPSCALE enlarges the latent video using 3D latent upscaling and performs a second sampling pass.

Each stage can use different models, Turbo LoRAs, enhancement LoRAs, strengths, sampling steps, samplers and schedulers.

This allows users to configure the first stage for generation speed and motion, while configuring the second stage for refinement and final visual quality.

Use Single or Chain mode

Single mode generates one standalone video segment.

Chain mode generates multiple connected segments and manages their execution, prompts and continuity states from the Director interface.

Monitor generation live

The integrated monitor displays generation progress and animated previews directly inside the Director.

It uses KJNodes Model Preview Override and listens to the kj_preview_override event, together with standard ComfyUI progress events.

ComfyUI-KJNodes is required for animated live previews. A separate queue-monitor node is not required.

Keep the workflow flexible

The Director does not lock the user into a completely closed graph.

External connections remain available for:

• Image, video and audio references

• Alternative model loaders

• Different LoRA configurations

• Compatible latent upscalers

• Video and audio export

• Additional ComfyUI processing nodes

• Custom experimental configurations

Main features

• Centralized MiniMax H3 Director interface

• BASE generation and second UPSCALE sampling pass

• Integrated 3D latent upscaling

• Configurable upscale multiplier

• Single and Chain generation modes

• General Prompt and independent segment prompts

• Add, expand, collapse and select video segments

• MiniMax H3 Motion Context integration

• Save and restore continuity states

• Automatic context trimming

• Independent BASE and UPSCALE models

• Independent BASE and UPSCALE LoRA stacks

• Independent sampling settings for both stages

• Configurable duration, resolution and aspect ratio

• Fixed or randomized seed

• Preview BASE and Final modes

• External image, video and audio references

• Integrated animated live preview

• Final video export with audio

• Accessible external ComfyUI connections

Installation

Open the ComfyUI/custom_nodes directory and run:

git clone https://github.com/M1cr0Bytes/MiniMax-H3-Chain-Director-V1.git ComfyUI-H3-Director-Studio-V1

Alternatively, download the repository as a ZIP and extract it to:

ComfyUI/custom_nodes/ComfyUI-H3-Director-Studio-V1

The folder must directly contain:

__init__.py
web/studio.js

Do not leave the project inside an additional nested directory.

After installing the required custom-node packages:

1. Restart the ComfyUI server.

2. Reload the ComfyUI web interface.

3. Search for H3 Director Studio in Add Node.

4. Download the public workflow JSON from this Civitai page.

5. Import the JSON into ComfyUI.

6. Select the compatible models, VAEs, LoRAs and latent upscaler installed on your system.

Complete installation guide:

https://github.com/M1cr0Bytes/MiniMax-H3-Chain-Director-V1/blob/main/docs/INSTALLATION.md

Required custom-node packages

The public workflow uses components from:

• Deno Custom Nodes

• H3 Motion Context

• KJNodes

• H3 latent upscaler

• OBVPM Timeline

• rgthree-comfy

• VideoHelperSuite

• ComfyUI-Pixaroma

Complete dependency list and upstream links:

https://github.com/M1cr0Bytes/MiniMax-H3-Chain-Director-V1/blob/main/docs/DEPENDENCIES.md

Important dependency notes

• Deno Custom Nodes supplies the MiniMax H3 reference-loading components used by the Director.

• H3 Motion Context provides the continuity system used in Chain mode.

• KJNodes is required for animated previews in the integrated live monitor.

• The compatible H3 latent upscaler is used by the Final/UPSCALE branch.

• rgthree-comfy provides the two external LoRA stacks used by the public workflow.

• VideoHelperSuite provides the final video and audio export node.

• PixaromaRunTimer only measures generation time. It is not required for the Director’s underlying video generation, but Pixaroma must be installed to open the supplied JSON without its missing-node warning.

• Exact minimum-compatible versions have not yet been pinned.

Models and LoRAs

Models, VAEs, latent-upscaler weights and LoRAs are not included.

Users must separately install:

• A compatible MiniMax H3 diffusion model

• MiniMax H3 text encoder

• Video VAE

• Audio VAE

• Compatible H3 latent-upscaler checkpoint

• Any Turbo or enhancement LoRAs selected in the workflow

Developer’s tested configuration

The public workflow is configured with an emphasis on photorealism, natural textures, realistic motion and fine visual detail.

Recommended BASE options:

• Singularity Ref2VA v1.3

• 10Eros TURBO-hybrid Beta5

Preferred UPSCALE configuration:

• MiniMax H3 FL2V Turbo 4-Step

• H3 LMS latent upscaler — h3_upscaler_lms_v0.1

These are practical recommendations based on the developer’s tests and visual preferences. They are not mandatory dependencies or a universal quality ranking.

Compatible alternative models, LoRAs and upscalers can be used for realistic, cinematic, stylized or artistic results.

The example workflow applies the optional Attention acceleration backend only to the UPSCALE stage.

Recommended first test

After installation, use a short clip and your own authorized references:

1. Run Single + BASE Preview and confirm that frames and audio are generated.

2. Run Single + Final and confirm that the latent upscale and second sampling pass complete.

3. Run Chain with two segments and verify segment progression and continuity.

4. Check the transition frames, dialogue timing and audiovisual synchronization.

5. Test different BASE and UPSCALE model or LoRA configurations independently.

Hardware, ComfyUI version, models and third-party node versions can affect performance and results.

Public workflow

The supplied workflow is a sanitized public example.

It does not include private prompts, personal reference files, local model weights or third-party LoRAs. Model and LoRA selections may need to be changed to match the filenames installed on your system.

The public graph contains the Director and the required external connections, allowing users to inspect and adapt the workflow instead of being locked into an inaccessible internal configuration.

Feedback and issue reports

Feedback from different ComfyUI installations is welcome, especially regarding:

• Installation and missing dependencies

• BASE Preview

• Final/UPSCALE generation

• Chain execution

• Motion Context continuity

• Transition trimming

• Audio preservation and synchronization

• Integrated live preview

• Different model and LoRA combinations

When reporting a problem, please include your ComfyUI version, GPU, selected models, custom-node versions and the relevant console error.

License and credits

The Director code is released under the MIT License.

Third-party custom nodes, models, VAEs, LoRAs and upscaler weights retain their original licenses and terms.

This project is not affiliated with or endorsed by MiniMax, ComfyUI or the authors of the upstream dependencies.

Credits and acknowledgements:

https://github.com/M1cr0Bytes/MiniMax-H3-Chain-Director-V1/blob/main/docs/CREDITS.md

The Director provides the tools. The creative choices belong to the user.