Ino S3 Sync Folder

The other S3 nodes move single objects. Ino S3 Sync Folder moves everything - a whole local folder up to S3, or a whole bucket prefix down to disk, with concurrency, change detection, and a full accounting of what happened. It's the pack's answer to "I don't want to upload 200 images one by one." If you're backing up a run's outputs or restoring a job's inputs on a fresh cloud box, this is the node you want.

It's part of ComfyUI Ino Nodes (nobandegani/ComfyUI-InoNodes), the 125+-node V3-schema pack, and the heaviest of its 16 S3 helpers. It's also an output node, so it runs even if nothing consumes its results - which is what you want for a "do the sync, then tell me about it" tool.

How it works

The key input is sync_local, which picks the direction. Despite the name, on means download (S3 → local), off means upload (local → S3) - the toggle label spells it out: "Download (S3->local)" / "Upload (local->S3)". Default is on (download).

Other inputs:

• enabled - the pack-wide toggle.
• s3_key - the bucket prefix (folder) to sync with.
• parent_folder / folder - the local side, via the standard input/output/temp dropdown plus subfolder (default sync/).
• s3_config - optional; config string or env vars.
• concurrency - how many transfers run in parallel (1–10, default 5). Higher is faster on many small files, but harder on the network.
• delete - the dangerous one, and the tooltip is explicit: if on, orphan files on the destination side get removed, rsync-style. Defaults off. Only turn it on when you deliberately want one-way mirroring - it's the difference between "make them match" and "nuke whatever isn't in the source."

Outputs: success, message, rel_path, abs_path, then a full scoreboard - downloaded, uploaded, skipped_unchanged, failed, removed_local, removed_remote, and errors (a string of what went wrong). The skipped-unchanged count is the nice touch: the sync detects files that already match and skips them, so re-running the same sync is cheap.

When you'd use it

Two patterns dominate: upload at the end (ship an entire output folder to S3 for archival or delivery) and download at the start (restore a job's inputs from the bucket onto a fresh machine - the classic cloud-GPU move). The change detection means you can re-run the sync as a safety net without re-transferring everything.

Common issues

• delete is a loaded word. Leave it off until you're sure. With it on, files on the destination that aren't in the source get deleted - there's no trash can in S3 or in your folder.
• Direction confusion. sync_local on = download. Everyone reads this backwards once; the toggle label is your friend.
• Counts as your source of truth. If failed is non-zero, don't assume the sync completed - check errors.

Installing it

ComfyUI Ino Nodes installs like any custom node pack:

• ComfyUI Manager (easiest): search for "ComfyUI Ino Nodes", install, restart.
• Manual:

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

  Then restart ComfyUI.

Requires inopyutils and a current ComfyUI. It's a niche pack with almost no Reddit footprint - the S3-with-ComfyUI pattern itself gets plenty of discussion, but this pack rarely shows up in it. For folder-sized transfers to object storage, though, it's the most complete tool in the pack.