MiniMax H3 ComfyUI workflow and local setup: where to start

Mind Forge AI editorial team · Content updated: 2026-11-10

Start with the MiniMax H3 templates maintained by Comfy-Org and a matching checkpoint from its model repository. If you only want to try a short video without managing a GPU, Mind Forge AI also offers a daily-limited free web preview. That preview is silent SD video; it does not expose every capability of the wider H3 family.

Choose the tool and check its scope

AI Image & Video Models

Explore the model choices available in Mind Forge AI before starting an image or video task.

Availability, input support and credit cost vary by model. The workspace shows the current options before submission.

How to prepare a MiniMax H3 ComfyUI workflow

Choose the workflow before downloading weights. Text/first-frame generation uses the fl2va family; reference-to-video uses ref2va. The template's node and checkpoint names must match.
Install or update ComfyUI using its documentation. Check the model card and template for the required diffusion model, text encoder and VAEs. Place files in the corresponding ComfyUI model directories and refresh the model list.
Import the template, select the installed files and use a short, low-resolution test first. Check execution errors before increasing duration or resolution. These are setup steps, not a promise that a particular GPU can run the selected profile.

A prompt you can adapt

A ceramic cup on a wooden desk by a window. Locked camera, soft morning light and a small ribbon of steam rising slowly. Keep the scene simple and the whole cup visible.

This is a suggested prompt, not a verified generated result.

A short test before committing to a cloud GPU

This is a workflow example, not a completed generation case.

Prepare the source

Use a simple text scene with no reference media. Record the chosen workflow, checkpoint, duration and output size so that a later change has a clear baseline.

Choose the workflow

Check the official template and current model terms first. For cloud use, set a spending limit, account for persistent storage and shut down idle compute when finished. A no-deployment web preview can help you decide whether to continue exploring video generation.

Correct problems

If a node is missing, verify the ComfyUI version and the template's dependencies. If memory is insufficient, stop the failed test and review the checkpoint and output settings before retrying; a free-weight download does not guarantee a low-cost run.

Just want to try H3 without deployment?

Use the browser-based free preview before setting up a local workflow or renting cloud compute. It has daily limits and outputs silent SD clips; reference-audio and multi-image workflows are not enabled in this form.

Explore the free PinkCherry H3 workflow

Workflow downloads, weights and setup resources

These links point to external maintainer resources, not the site's private workflows, service credentials or user tasks. Check files and current terms for your selected version.

Review your result

Check node availability, checkpoint filenames, input types and the selected output settings. A reference workflow is not interchangeable with a first-frame workflow. Model files are distributed under their stated terms; access to weights does not include free GPU time, storage or bandwidth.

Costs and limits

Guides are free to read. The dedicated MiniMax H3 and PinkCherry H3 free video previews need no sign-in or credits and have daily limits; they produce silent SD clips. Other paid creation tools require sign-in and sufficient credits; review the quote before submission. Image and video generation are separate tasks. Confirmed paid-task failures return the original task credits; pending or unsatisfactory completed outputs are not confirmed failures.

Do I need to deploy ComfyUI to use free MiniMax H3 online?

No. The website's free preview runs through a browser form and does not require downloading weights or installing ComfyUI. Choose the local or cloud route when you need to manage the workflow yourself, after checking model terms and compute costs.

Continue creating

Open the relevant tool →