Built by ZEM · Made for ComfyUI

Custom nodes.
Ready-to-use workflows.

I create custom ComfyUI tools that make complicated setups easier to use. Start with the ZEM H3 Multi-Image Adapter: add your reference pictures, describe your video, and use the included workflow to run it.

This workflow requires the custom ZEM adapter.

The included workflow only works because the ZEM H3 Multi-Image Adapter handles its reference images. The online generator already has the adapter installed. To run the downloaded workflow in your own ComfyUI setup, install the custom node pack below first; otherwise, ComfyUI will show a missing node and cannot run this workflow as supplied.

01 / THE RECIPE · JSON

Download the workflow

The saved ComfyUI layout, with all nine image connections already in place. Open this file in ComfyUI after installing the adapter.

Download workflow (.json)

02 / THE EXTRA TOOL · ZIP

Download the custom node pack

The ZEM adapter, its upload controls, installation notes, and a copy of the workflow. Extract it into your ComfyUI custom_nodes folder.

Or install with Git →

Download node pack (.zip)

Already deployed — use this workflow online.

This workflow is already running on my site, powered by the custom ZEM adapter. Try the Standard Reference Video Generator (R2V) with up to 10 generations. Upload your pictures and create a video without installing ComfyUI or downloading the node pack. Pro access is unlocked separately through the AI Toolbox password.

Try Standard R2V → 10 uses

What does this plugin actually do?

Think of a workflow as a recipe and a node as a tool used in that recipe. This plugin adds one tool: an image-upload panel for the local MiniMax H3 video generator. It gathers your pictures and passes them to the generator in the right order.

Use one to nine still pictures. Each slot has upload and clear controls plus a preview, so you can see which pictures you selected.
Leave spare slots empty. The adapter passes empty values that a compatible H3 node skips. You do not need filler pictures or duplicate images.
Refer to pictures in your prompt. Click a displayed tag, such as <Picture 1>, to copy it. For example: “Use the person in <Picture 1> in the setting shown in <Picture 2>.”
Skip the manual wiring. The included workflow already connects all nine outputs to H3; the adapter hides the connection sockets to keep the panel simple.

Why you need it: this particular workflow uses ZEM's upload panel instead of separate image-loading boxes. The adapter is the part that turns those uploads into the inputs H3 expects. It makes changing the number of reference pictures easier; H3 and your installed models still do the actual video generation.

Install with Git

Already installed the ZIP? Skip this Git installation.

Choose one installation method only. Running the clone command after installing the ZIP is unnecessary and can cause a folder conflict or duplicate copies of the adapter.

Open a terminal on the computer or pod running ComfyUI. Go to its custom_nodes folder, then clone the pack. This example uses a common pod path; replace it with your actual path.

cd /workspace/ComfyUI/custom_nodes
git clone https://github.com/AlienZEM/zem-h3-multi-image-adapter.git

Already installed with Git? Pull the latest changes, then restart ComfyUI and refresh its browser page:

git -C /workspace/ComfyUI/custom_nodes/zem-h3-multi-image-adapter pull

View the source and README on GitHub →

Prefer the ZIP?

Extract the node pack and place the zem-h3-multi-image-adapter folder inside ComfyUI/custom_nodes. Check that __init__.py and nodes.py sit directly inside that adapter folder, not inside a second nested folder. Choose either Git or ZIP installation so you do not install duplicate copies. Git updates require a Git installation.

Run your first workflow

Check your setup. You need ComfyUI with local MiniMaxH3ReferenceToVideo support, the models used by this workflow, and hardware capable of running them. The adapter does not download model weights or add extra pip dependencies.
Install the node pack using Git or the ZIP instructions above. Restart ComfyUI and refresh the browser page.
Open the downloaded workflow JSON in ComfyUI. Check that the model selections match your installed files.
Upload at least one still picture in the ZEM H3 Multi-Image Adapter panel and wait for the upload to finish.
Write your prompt using the tags beside your pictures, then queue the workflow. Start with one image to check your setup before adding more.

A few things to know

Picture numbers follow the occupied slots from left to right. If you fill slots 1, 3, and 9, their tags are Picture 1, 2, and 3. Clearing a picture renumbers later tags; update your prompt yourself because the plugin does not rewrite it.
Use still images. Animated GIFs aren't supported. Upload at least one valid image; empty input or invalid files stop with an error.
This adapter is for the local MiniMax H3 node. It does not work with the MiniMax partner/API node or arbitrary image nodes, and it does not configure a hosted endpoint.
If H3 errors on empty references, update ComfyUI to a version whose local H3 node skips empty inputs. The package has automated checks; a live ComfyUI browser session and GPU generation have not been verified for this release.