ZhipuAI VideoReportData

After VideoReportGenerate and VideoReportPull do their thing, you're holding absolute paths to CSVs full of task IDs, Chinese prompt expansions, and hosted video URLs - and zero way to see any of it inside ComfyUI. That's the entire job of VideoReportData. It's the least glamorous node in the Zhipu pack and also the one that makes the other two usable, because a pipeline whose outputs you can't inspect is a pipeline you don't trust.

How it works

It's almost aggressively simple: give it a file_path, and it opens the file with Python's plain csv.reader, then returns the rows as a list of lists. No header magic, no schema sniffing, no column remapping. In this pack that's a feature rather than a shortcoming - the reports it's meant to read are already clean tables the pack generated itself, so a dumb dump is exactly what you want.

Two outputs fall out of that single read: data (DICT, the raw rows) and show_text (STRING, the same rows flattened to text). The show_text output is the one you'll actually wire somewhere - in the pack's example workflow it feeds a ShowText node so you can eyeball the pull report, see which videos finished, and grab their URLs. data exists for any future node that wants the rows programmatically; today nothing in the pack consumes it.

The one input worth knowing about:

• file_path - a STRING. It's marked as a fork input, which means you can connect a path from another node rather than typing it, and that's the right way to use it. Wire VideoReportPull's video_pull_report_path output straight in and the loop is closed: submit, poll, read.

Installing it

Same pack, same install as its siblings. ComfyUI Manager, search ComfyUI ZhipuAI Platform, install, restart. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/MetaGLM/ComfyUI-ZhipuAI-Platform

No model downloads - everything runs against Zhipu's hosted API, and this node in particular doesn't even call it. It's a local file read. The only environment requirement you need to inherit from the rest of the pack is the API key in whatever shell launches ComfyUI (export ZHIPUAI_API_KEY=...), plus the dependency's Python 3.10/3.11 pin. Note the read is local, so it works even if you never set the key - handy for inspecting a report saved by an earlier session.

Where people get burned

• Wrong path, ugly error. Point it at a file that doesn't exist and you get a raw FileNotFoundError traceback, not a polite message. Double-check absolute paths, or keep them wired from the previous node.
• It's not a general CSV tool. If you feed it an arbitrary spreadsheet expecting headers to become column names, you'll get rows back exactly as stored - headers included, as row zero. This node is tuned for the pack's own reports, and it will not think for you.
• Empty-looking output is a timing issue, not a node issue. If show_text comes back as [[]] or an empty list, you're reading a report that has no finished rows yet - usually because the video is still rendering and Pull hasn't collected results. Re-run the pull step, don't hunt for a bug in the reader.

It's a tiny node that does one thing with no cleverness, and honestly that's the right design for the last mile of an async pipeline. When your Generate→Pull chain works and you can finally read your video URLs inside ComfyUI, this is the quiet node you have to thank.