MiniMax H3 Easy Media Bridge

The one node you only need if you're automating ComfyUI

The main MiniMax H3 Easy node accepts multiple images, videos and audio clips through a single multi-link Media port. It's a nice way to work on the canvas, but it has a secret: that port is a frontend illusion. A web extension turns your visible wires into hidden execution inputs and remembers their order. On the canvas that's elegant. Through the ComfyUI API, or a headless runner, or any server-side executor, there is no canvas and no extension - the virtual port has nothing to back it.

MiniMaxH3EasyMediaBridge is the escape hatch. It collects media through explicit, boring, API-friendly inputs and bundles them into a MINIMAX_H3_MEDIA_BUNDLE you can feed into the main node's Media port. If you're scripting workflows or running them server-side, this is the node that makes the whole pack usable.

Inputs

Three counts, then a bank of numbered sockets:

• image_count - 0 to 9. Controls how many image_N inputs appear.
• video_count - 0 to 3.
• audio_count - 0 to 3.

Each count toggles its numbered inputs on: image_1…image_9, video_1…video_3, audio_1…audio_3, all optional wildcards. You set the counts to match your source, connect each media node to its matching numbered input, and you're done. Unused sockets simply don't get passed along.

Output

• media_bundle (MINIMAX_H3_MEDIA_BUNDLE) - wire this into the main Easy node's Media port. The main node recognizes a media bundle directly and unpacks it, keeping each item's type and order. That's the entire contract.

For normal canvas workflows, ignore this node and connect media straight to Media - the bridge buys you nothing there and adds a node to read around. It exists for the API path.

Install

Same as the rest of the pack:

cd ComfyUI/custom_nodes
git clone https://github.com/nkxx188/ComfyUI-MiniMaxH3-Easy

or search "MiniMax H3 Easy" in ComfyUI Manager, then restart ComfyUI. No extra dependencies; it's pure plumbing over the pack's own bundle type.

Common issues

The classic mistake is leaving the counts higher than the inputs you actually feed. The node quietly skips None inputs, so a video_count of 3 with only one video connected just produces a two-item bundle - but if you expected the main node to see three items, your numbering downstream will be off. Set counts to match reality. Second: this node packs, but it doesn't validate against mode limits - the main node still enforces "max 9 images, 3 videos, 3 audio" and "reference mode needs at least one image or video" when it runs. Third, remember the counts are per type and numbering restarts for each type; the main node's @ references and <Image N> tags follow that same independent numbering, so keep the bridge's order aligned with how you write the prompt. If the two disagree, you'll get references pointing at the wrong media rather than an error.