Skip to main content
AI tutorials

Generate Videos with Cloud GPUs Using MiniMax H3 Cloud in Codex

Learn to use MiniMax H3 Cloud for Codex to generate videos with CompShare cloud GPUs while securely managing credentials, costs, and instance shutdowns.

About 20 min read
Generating videos with MiniMax H3 Cloud in Codex

Project Overview

MiniMax H3 Cloud for Codex is a cloud GPU-powered video generation plugin. It lets users submit requests directly in Codex, while the plugin handles compute selection, instance creation or startup, ComfyUI Runtime preparation, workflow submission, MP4 downloads, and instance shutdown after the task ends.

The plugin currently supports the CompShare platform, and the README states that additional platforms will be added in the future. The plugin itself is free, but cloud resources such as GPUs, system disks, and images are billed directly by CompShare.

License notice: The repository does not include MiniMax H3 model weights. The model is governed by the MiniMax H3 Community License Agreement, which currently explicitly excludes regions including the European Union, the United Kingdom, South Korea, and the United States. Users in excluded regions must first obtain separate authorization from MiniMax and confirm before use that their location, intended use, and platform services comply with the applicable terms.

Core Features

Three Video Generation Modes

  • Text-to-video: Generate a video from a text prompt.
  • Keyframe-to-video: Provide a first frame, a last frame, or both.
  • Reference-image-to-video: Provide 1 to 9 reference images.

Automated Cloud Workflow

The plugin helps you check GPU prices and availability. After you confirm the configuration, it handles the remaining process, including starting a compute instance, preparing the ComfyUI Runtime, submitting the video workflow, downloading the result, and shutting down the instance.

Cost and Shutdown Protection

The plugin must obtain user confirmation before creating paid resources. By default, it configures a 30-minute fallback shutdown and immediately shuts down the instance after an individual task succeeds or fails, reducing the risk of additional charges from an instance that continues running.

Local Credential Management

The plugin does not ask users to send their API Key in a Codex conversation. CompShare credentials are securely stored locally by the CompShare CLI. Plugin configuration and generated videos are also stored outside the plugin installation directory, so plugin upgrades will not overwrite this user data.

Prerequisites

Before installation, make sure your local environment meets the following requirements:

  • The operating system is macOS or Linux.
  • A version of Codex that supports plugins is installed.
  • Node.js 20 or later is installed.
  • ssh and ffprobe are available on the system.
  • You have a CompShare account and API Key.
  • CompShare CLI 0.3.5 or later is installed.

Install CompShare CLI

Install or upgrade CompShare CLI through PyPI, then check its version:

python3 -m pip install --upgrade compshare-cli
compshare --version

Confirm that the reported version is 0.3.5 or later. For the CompShare CLI project and additional information, see the CompShare CLI GitHub repository.

Install the MiniMax H3 Cloud Plugin

First, add the project repository to Codex Marketplace, then install the corresponding plugin:

codex plugin marketplace add Sac-Y/MiniMax-H3-Cloud --ref main
codex plugin add minimax-h3-cloud@sac-y-minimax-h3

After the commands finish, you can invoke MiniMax H3 Cloud in Codex. On the first run, the plugin guides you through account credentials, GPU configuration, and video parameter settings.

First-Time Workflow

  1. Sign in to the cloud platform: Register for or sign in to CompShare, then use the official CLI to securely store your API Key locally. Do not paste the API Key into a Codex conversation.
  2. Select a GPU: Check real-time GPU prices and availability, then choose a configuration that meets your requirements and budget.
  3. Provide generation parameters: Submit a prompt, provide images when required by the generation mode, and set the aspect ratio, duration, and resolution.
  4. Confirm the cost: Review the video parameters and estimated cost. The plugin creates paid resources only after you confirm.
  5. Wait for generation: The plugin automatically prepares the runtime environment, executes the workflow, and downloads the MP4.
  6. View the result: When the task is complete, you can play the video in Codex and download the MP4 file.

Basic Usage Examples

The following are examples of requests you can make in Codex. Refer to the guidance displayed during the plugin's first run for the exact options and available configurations.

Example 1: Text-to-Video

Use MiniMax H3 Cloud to generate a video of the seaside at dusk. The camera should move slowly forward, with gentle waves and an overall cinematic, warm color palette. Guide me through choosing the aspect ratio, duration, resolution, and GPU.

The plugin enters the parameter collection and GPU selection process based on your request, then displays the estimated cost before creating resources.

Example 2: Using a First and Last Frame

Use MiniMax H3 Cloud to generate a video from the first and last frames I provide. Keep the subject consistent, and let me confirm the video parameters and estimated cost before execution.

Keyframe mode supports providing only a first frame, only a last frame, or both a first and last frame.

Example 3: Using Multiple Reference Images

Use MiniMax H3 Cloud to generate a video from the reference images I provide, and guide me through configuring the aspect ratio, duration, resolution, and compute resources.

Reference-image-to-video mode supports 1 to 9 reference images. Before submitting the task, confirm that the number of images is within the supported range.

Configuration and Output Locations

By default, the plugin writes its configuration to:

~/.config/minimax-h3-cloud/config.json

Generated videos are saved by default in:

~/.local/share/minimax-h3-cloud/outputs/

Both directories are outside the plugin installation directory, so plugin upgrades will not overwrite your configuration or generated videos. To create a backup, copy the relevant files from these directories separately.

Advanced Usage and Important Considerations

Compare Real-Time Prices and Availability First

Cloud GPU prices and availability change in real time. Before starting a task, choose a configuration based on your budget, available inventory, and video requirements. If no qualifying resources are available, the plugin returns to the GPU selection step instead of silently switching to a configuration you have not approved.

Review Paid Resource Confirmation Details Carefully

The plugin itself is free, but the cloud platform charges for resources such as GPUs, system disks, and images. Before confirming resource creation, verify the selected GPU, estimated cost, and video parameters to avoid unnecessary expenses caused by an incorrect configuration.

Use Automatic Shutdown to Control Costs

The plugin provides a 30-minute fallback shutdown by default and immediately shuts down the instance after an individual task succeeds or fails. Even so, you should monitor the task status and the cloud platform's resource page to confirm that the instance has stopped as expected.

Do Not Expose Your API Key in Conversations

Your API Key should be stored locally through CompShare CLI and must not be shared in Codex conversations, logs, screenshots, or public issue reports. The plugin's normal workflow never asks you to provide credentials in chat messages.

Choose a Generation Mode Based on the Input Type

  • Use text-to-video when you only have a creative description.
  • Use keyframe-to-video when you need to control the opening or closing image.
  • Use reference-image-to-video when you need multiple visual references, and keep the number of images between 1 and 9.

Review the Model License Before Use

The plugin code uses the MIT License, but the model and workflows have their own licensing requirements. In addition to reviewing the MiniMax H3 Community License Agreement, read the repository's third-party licensing information and confirm that your region and intended use comply with the requirements.

Frequently Asked Questions

Does the plugin include MiniMax H3 model weights?

No. The README explicitly states that the repository does not provide MiniMax H3 model weights.

Can it be used on Windows?

The README lists macOS and Linux as supported environments and does not state that Windows is supported.

Which cloud platforms are currently supported?

CompShare is currently supported, and the project plans to add other platforms in the future.

Will it automatically switch to another configuration if a GPU is unavailable?

No. If no qualifying resources are found, the plugin returns to the GPU selection step and does not switch to another configuration without confirmation.

Where are generated videos saved?

By default, they are saved in ~/.local/share/minimax-h3-cloud/outputs/.

Conclusion

MiniMax H3 Cloud for Codex integrates cloud GPU selection, ComfyUI Runtime preparation, video workflow execution, MP4 downloads, and instance shutdown into the Codex conversational workflow. After installing the dependencies and configuring local credentials, users can generate videos from text, keyframes, or 1 to 9 reference images. In actual use, pay close attention to the model license, real-time resource pricing, paid resource confirmation, and API Key security.