Chat GLM Video Generate Node

Local video generation (Wan, LTX) is heavy - long generation times, real VRAM demands, a whole workflow to get right. ChatGLMVideoGenerateNode sidesteps all of that by calling Z.ai's hosted CogVideoX models over an API instead: send a text prompt (optionally with a starting image), and a finished video comes back. No local weights, no sampler steps to tune, no GPU time spent on your own machine.

Why you'd reach for it

CogVideoX is genuinely a big deal historically - it was one of the first credible open competitors to Kling and Runway back in 2024, and Z.ai (formerly Zhipu AI, the Tsinghua-spinout lab behind it) has kept iterating on it even as Wan became the community's default for local open video generation. This node gets you the hosted version of that model line, which matters if you want a quick video test without setting up a local video pipeline at all, or you're on hardware that simply can't run Wan or LTX locally.

The trade is the same as every other API node in this pack: you're paying per generation instead of paying once in setup time and GPU-hours, and you're working within whatever parameters Z.ai's hosted endpoint exposes rather than the full knob-set of a local ComfyUI video graph.

How it works

Your prompt (and optionally a starting image, for image-to-video) go to Z.ai's API, which runs a CogVideoX model server-side and returns a finished video file. There's no local sampling, no VAE decode step on your end - the whole generation happens remotely.

The inputs and outputs that matter

• prompt - your text description, multiline.
• model (default cogvideox-flash) - 3 choices: cogvideox-3, cogvideox-2, cogvideox-flash. Flash is the fast/cheap default; the numbered variants are presumably higher-fidelity, slower options.
• image (optional, IMAGE) - feed a starting frame for image-to-video generation instead of pure text-to-video.
• quality (default speed, optional) - speed or quality, the straightforward trade-off.
• with_audio (default false, optional) - whether the generated video includes audio.
• watermark (default true, optional) - turn off if you need clean output (check Z.ai's terms before relying on watermark-free output commercially).
• width / height (default 1920×1080, optional) - output resolution.
• fps (default 30, optional) - 30 or 60.
• duration (default 5, optional) - 5 or 10 seconds.

Output is a single VIDEO - wire it into a Save Video node or further processing.

How to install it

Via ComfyUI Manager: search "ComfyUI Custom Nodes AlekPet" and install. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/AlekPet/ComfyUI_Custom_Nodes_AlekPet

Restart ComfyUI. You'll need a Z.ai (Zhipu) API key configured before this generates anything - check the ChatGLMNode subfolder in the repo for exactly where the key goes, since the top-level README doesn't document per-node API setup. No local model download; generation happens entirely server-side.

Common issues & troubleshooting

Authentication errors. Unconfigured API key - every ChatGLM* node in this pack needs one, and video generation is no exception.

Cost adds up fast compared to local generation. Video API calls are typically the most expensive tier of hosted generation, more so than the image or text nodes in this pack. If you're iterating on a prompt, expect that to show up on your bill quickly - this node is better suited to a final render or a one-off test than to rapid iteration.

Generation takes noticeably longer than the image nodes in this pack. Expected - video generation is inherently heavier than a single image even on well-provisioned server hardware. Don't assume a hang; give it time, especially at quality mode or the 10-second duration option.

Output resolution/duration doesn't match what you requested. Double-check the hosted endpoint actually supports your combination of width/height/fps/duration - hosted APIs sometimes silently clamp to supported presets rather than erroring, so if something looks off, try one of the schema's listed default combinations (1920×1080, 30fps, 5s) as a known-good baseline before troubleshooting further.