arkennemasis Catalogue Fan-Out (one chain, N rows, concurrent)

ArkCatalogueFanOut is the loop of the catalogue path. You build one generation chain - image edit node, save node, whatever the run needs - and this node makes that single chain render every row of the catalogue, concurrently. No copying the chain N times, no for loop in your head, no 200-node canvas. Wire it between ArkCatalogueLoad and your generator and the fan-out is the plumbing.

The mechanism is ComfyUI's OUTPUT_IS_LIST: everything this node emits except count is a list, and ComfyUI responds by running the downstream chain once per item. That's what turns one graph into N executions. Because the rows run through the same chain, the image generator's own concurrency gate overlaps the calls - which is the polite way of saying you get parallelism without blasting a rate limit.

How it works

The only required input is rows_json, straight from ArkCatalogueLoad's rows_json output. The node reads the rows, loads each row's model and cloth reference images from disk, and fans everything out:

• model_image → wire into the generator's image_1
• cloth_image → image_2
• prompt → prompt
• filename → the save node (the CSV's own filename, so files land where the spreadsheet says)
• sku, index - per-row identifiers for naming and records
• count - not a list; the total, for nodes that need to know how many there are

If there's nothing to generate - every row already delivered, or the filter on Load matched no rows - it raises and tells you so, rather than silently doing nothing.

Installing

cd ComfyUI/custom_nodes
git clone https://github.com/Hishamahmer/comfyui-arkennemasis
pip install replicate httpx

Restart ComfyUI. It's pure list plumbing - no models, no keys.

Troubleshooting

• "nothing to generate" - all rows delivered, or the filter matched nothing. Check the Load node's only/skip_existing settings rather than assuming a broken graph.
• One row per chain run means one failure can be per-row. The generator's retry logic (in the pack's API nodes) handles transient 429s and 5xx; a hard failure on one row aborts that run. That's why skip_existing on Load matters - a failed run resumes without re-billing what already succeeded.
• Concurrency caution: the pack's API nodes default to running one call at a time with an option for all at once (capped at 2 by default). If you're paying per call, keep an eye on which mode your generator is in before you fan out 60 rows.