1.5.Grok 批量提交 (Wujiai)

The single Grok submit node is fine for one clip. This is the one for "I have 40 prompts in a spreadsheet and I want 40 videos." GrokSubmitBatchTask sits at "1.5" in the flow - after the single submit (1), before the status query (2) and download (3) - because it's just the batch flavor of step 1. Same channel, same async pipeline, same query/download nodes downstream; it just fires many jobs at once.

Which is exactly why you should think before using it. Video generation is the expensive end of the cloud-model market - the Wujiai channel meters per call, and a batch node is a machine for spending credits at scale. The tool itself is solid: it reads prompts from a CSV/Excel column, optionally attaches reference images to every job, and fires them with a bounded concurrency. But a batch of 15-second videos adds up fast. That's not a bug, it's the business model.

The inputs that matter

• file_path - path to a .csv, .xls, or .xlsx. This is where your prompts live.
• column_name (default prompt) - which column holds the prompt text. The obvious beginner trap: if the name doesn't match a column exactly, every job fails.
• aspect_ratio - 2:3, 3:2, 1:1, 9:16, 16:9 (default 16:9).
• duration - 6, 10, or 15 seconds. Remember: this applies to every row.
• concurrency (INT 1–10, default 2) - how many jobs it fires in parallel. Keep this humble; it's the difference between "30 requests over a few minutes" and "30 requests in one burst."
• images (optional IMAGE) - reference frames attached to all submissions in the batch (image-to-video for every row).
• api_key (optional) - falls back to the pack-level Wujiai key.

Outputs are task_ids (STRING) and report (STRING) - the receipts you'll cross-reference in GrokQueryTasks.

Install

One pack, one install:

cd ComfyUI/custom_nodes
git clone https://github.com/jieg9341-lab/ComfyUI-nkxx

restart (or ComfyUI Manager, "ComfyUI-nkxx"). Auto-installs requests, pandas, openpyxl, yt-dlp, opencv-python, aiohttp on first launch; no models download because nothing runs locally. You need a Wujiai key from wujiai.org.

Common issues

• "列 'x' 不存在" (column not found): your CSV header doesn't match column_name. Open the file, check the exact header.
• Credit shock: test with 2–3 rows and duration: 6 before unleashing the whole sheet. concurrency protects throughput, not your wallet.
• Partial failures: batch nodes report per-task status in report; a few failures don't stop the rest, so read the whole report, not the first line.
• Trust note: the Wujiai-side nodes ship obfuscated in this pack. The batch logic is only verifiable by behavior - run a tiny batch first and check the billing line on the Wujiai dashboard.