Ino S3 Download Folder

Some workflows don't need one file from S3 - they need the whole folder: a dataset, a model bundle, a run's output archive. InoS3DownloadFolder is the bulk version of the pack's S3 downloader, and it's built like a proper sync tool rather than a naive loop. It downloads an entire folder from a bucket with concurrent transfers (default 8, up to 16), skips files that already match locally unless you force overwrite, and can verify the result afterward. You get a full report on the other end: totals, successes, skips, failures, and the error strings for anything that didn't make it.

That reporting is the difference between "it worked, I think" and "123 files synced, 2 skipped, 0 failed." If you're doing the cloud-GPU dance - upload a dataset to S3, spin up a rented box, pull the data down, push results back up - this node is the pull half of that loop.

How it works

Required inputs: enabled, s3_key (the folder prefix in the bucket - e.g. input/ or datasets/faces/), parent_folder, and folder (the local destination subfolder).

Optional inputs worth understanding:

• max_concurrent - parallelism for downloads, default 8.
• verify - if True, checks the folder is fully synced after download. The tooltip notes it's skipped if any downloads failed, which is a sensible "don't certify a broken sync" guard.
• overwrite - default False; when False, each file is skipped if the local copy already matches the S3 object. Flip to True to force everything down fresh.
• s3_config - credentials JSON, or leave empty for the env vars (S3_ACCESS_KEY, S3_ACCESS_SECRET, S3_BUCKET_NAME, S3_ENDPOINT_URL, S3_REGION_NAME).

Outputs: success, message, rel_path/abs_path, then the report: total_files, downloaded_successfully, skipped_files, failed_downloads, and errors (the failure strings, so you don't have to guess).

Installing it

Part of the 125+-node ComfyUI-InoNodes pack by nobandegani:

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

Or search "ComfyUI Ino Nodes" in ComfyUI Manager, install, restart. V3 schema - current ComfyUI required.

Common issues

Folder downloads are where people first hit the "is this a key or a folder?" confusion. In S3 there are no real folders - s3_key is a prefix, so input/ with a trailing slash grabs everything under that prefix while input (no slash) might match objects named input-something. Use the trailing slash for the "everything here" semantics you want. Also, the skip logic is per-file: a 500-file folder with 5 changed files only downloads those 5, which is great until you wanted a clean re-download - that's what overwrite: True is for. If failed_downloads is nonzero, read errors; transient network failures on one object don't poison the whole run, and a retry with verify on is a decent reconciliation step.