Load Image From Minio

Normally the Load Image node in ComfyUI shows you files sitting in ComfyUI/input. Load Image From Minio does the same trick, except the "folder" is a bucket on a Minio server - possibly a different machine, possibly a different room. If you've built a pipeline where one box generates and another one post-processes, this is the node that feeds the second box.

The catch before you start

This node has exactly one input, image, and it isn't a text field - it's a dropdown of every file in your Minio input bucket, built at the moment the node is created or refreshed. That means two things. First, you must have run Set Minio Config successfully before you even add this node, or the dropdown comes up empty and the node errors out. Second, files you upload to the bucket later won't appear until you refresh the node - delete and re-add it, or force a refresh. Don't panic when your brand-new upload is missing from the list; it's just stale.

How it works

When the node runs, it grabs the config from minio_config.json (or env vars), fetches the selected object from the input bucket, and converts it exactly the way ComfyUI's core Load Image does: RGB pixels become a float tensor, and if the PNG has an alpha channel, that channel becomes an inverted mask. So the two outputs are:

• IMAGE - a standard image tensor, wires straight into VAE Encode or wherever your pipeline needs pixels.
• MASK - the alpha channel as a mask, same convention as the built-in loader. Useless if the source has no transparency.

Because it reuses the core loader's conventions, you can drop it into an existing workflow anywhere you'd use Load Image and it just works.

When you'd actually reach for it

Honestly? When you've committed to a shared-bucket workflow. Replacing your local loader with a remote one for the sake of it adds latency and a server to babysit for zero benefit - this is the kind of node that exists because of ComfyUI's single-box design, not despite it. Where it shines is in symmetry with Save Image To Minio: box A saves outputs to a bucket, box B loads them from its input bucket, and nobody is copying files over the network. It's the receiving end of a genuinely useful distributed setup.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/prodogape/ComfyUI-Minio
pip install -r requirements.txt

Or search "ComfyUI-Minio" in ComfyUI Manager, then restart. The only dependency is the minio Python SDK - one package, no models to download.

Troubleshooting

• Empty dropdown - Set Minio Config hasn't run (or hasn't succeeded) in this install. Run it first; it needs to write output/minio_config.json.
• "Failed to connect to Minio" - the pack's connection check is bucket_exists, and it never creates buckets. Make sure the bucket actually exists on the server.
• New files missing - refresh the node. The file list is read when the node's inputs are built, not on every run.