MiniMax H3 Workflow

Hey everyone!

Attached is our adapted version of the ComfyUI MiniMax H3 template. This workflow is very similar to the base template provided by Comfy, but adds a few additional elements identified by the community to help speed up and manage generation times.

Before you can use the workflow, you'll want to make sure your ComfyUI is up to date. If using the local version of ComfyUI, update it via the update_comfyui.bat file. If you're running in ComfyUI on Runpod, you can update ComfyUI from the manager inside of Comfy. If you're running in Runpod via our AI Influencer Runpod template, ComfyUI should already be updated for you.

Installing & Updating Custom Nodes

This workflows uses two different custom node packs, one you likely already have that needs to be updated, and another one that needs to be installed manually and cannot be found in the ComfyUI manager - let's start there.

Installing Spectrum Nodes

This process is pretty straight forward.

Step 1. Head to your ComfyUI folder then into "custom_nodes".

Step 2. In the navigation bar type "CMD" and hit enter. If running in Runpod, open a terminal window in your Jupyter Lab within the "custom_nodes" folder.

Step 3. Copy, paste, and enter the following line:

git clone https://github.com/xmarre/ComfyUI-Spectrum-MiniMax-H3.git

Step 4. Restart ComfyUI and refresh the ComfyUI page. The Spectrum node should no longer be missing.

Updating KJAI Nodes

KJAI recently introduced a node that allows us to preview the generation from a video model in real time. This is particularly useful for H3 which has quite long generation times.

Step 1. Head to your ComfyUI folder then into "custom_nodes".

Step 2. Find the KJAI node folder (if you have it) and click into it. If you don't have it, you can install them by loading the workflow into ComfyUI, clicking your "Manger" in the top right, the selecting "Install Missing Custom Nodes".

Step 3. In the navigation bar type "CMD" and hit enter. If running in Runpod, open a terminal window in your Jupyter Lab within the KJAI custom node folder.

Step 4. Run git pull to update to the latest version of the KJAI nodes.

Overall MiniMax H3 Thoughts

What immediately comes to mind is that this model is fairly slow, but at the quality we get, the results aren't half bad!

I think the overall conversation between LTX 2.5 and MiniMax H3 seems to land on the conversation of what you as a creator prioritize. If your priority is speed and pre-visualization, I personally think LTX 2.5 is a way better option. However, if you're looking for an option that is reliable with a consistent output quality and prompt adherence, MiniMax H3 seems to be a great option and a wonderful local source model compared to Seedance.