⏯️ RH Batch Execute

RH Batch Execute turns a parameter bundle into a pile of cloud tasks. You hand it a config, a workflow_id, and an RH_PARAM_BUNDLE (from RH Param Bundle, RH Batch Upload Image, or both), and it submits one task per job in the bundle, then returns all the task IDs as a single comma-separated string. That's the entire job: submit, collect IDs, hand them off.

Note the deliberate split. This node does not wait for results and does not download them. It fires tasks and gets out of the way. That's the pack being smart: a batch of ten video generations could run for half an hour each, and blocking your local queue on all of them would be miserable. Instead, Batch Execute returns fast, and you point RH Download at the returned task IDs whenever you're ready - it accepts comma-separated IDs and fetches them in parallel.

Inputs

• config - from RH Config. Required.
• workflow_id - the RunningHub workflow to run, pasted directly. Required, and separate from the config's workflow_or_app_id - yes, that's redundant-feeling, but it's how the node is built: the config holds your credentials, this field names the workflow for the batch.
• param_bundle - the RH_PARAM_BUNDLE of jobs. Required; an empty bundle raises "invalid or empty."

The single output is task_ids, a STRING of comma-separated IDs - one per submitted task, in bundle order.

How it works

It loops over the bundle, POSTing each job's params to /task/openapi/create (same endpoint RH Execute uses) with workflowId and nodeInfoList. Per-task failures are logged and skipped rather than aborting the whole batch - so if three of ten jobs have bad node IDs, you still get seven task IDs back. If all submissions fail, it raises.

Gotchas

The big one: this node spends real money fast. Ten jobs in the bundle is ten billed cloud runs the moment you hit Queue Prompt. There's no preview and no dry-run - if the bundle is wrong, you've spent the credits learning that. Relatedly, because it skips failed submissions silently-ish (console-only errors), check the console to confirm all N jobs actually landed; a batch that "succeeded" with 4 of 10 tasks is a confusing downstream surprise. And there's no built-in pacing - ten tasks hit the queue as ten simultaneous submissions, which can create a sudden queue jam (and occasional timeouts) on RunningHub's side.

Install: ComfyUI Manager (search ComfyUI_RH_API) or git clone https://github.com/xuchenxu168/ComfyUI_RH_API.git into custom_nodes + pip install -r requirements.txt, restart.