Kling Video Generation @fal

This node generates video with Kling - the text-to-video and image-to-video model - by calling fal.ai's hosted API. Unlike almost everything else in this pack, it does not run on your GPU. There's no model to download, no VRAM cost; it packages your prompt, ships it to fal's servers, and gives you back a URL to the finished video. Which also means it costs money: fal bills per generation, and you need your own fal API key. Read that twice before you wire it up - this is a paid cloud call wearing a ComfyUI node.

Why use it at all? Kling is a strong closed-source video model that you can't run locally, and this puts it inside a ComfyUI graph so you can drive it from the same workflow as everything else. If your bottleneck is "I want good video and I don't have the hardware for a local video model," a hosted-API node is a reasonable trade. If you'd rather keep everything local and free, this isn't your node.

How it works

You fill in a prompt (and optionally a starting image for image-to-video), pick the clip length and aspect ratio, choose a quality mode, and paste your fal key. On execution the node makes an authenticated request to fal's Kling endpoint, waits for the job, and returns the resulting video's URL as a string. The heavy lifting happens on fal's infrastructure; your machine is just orchestrating.

Inputs and outputs

• prompt (STRING, multiline) - the text description of the video.
• duration (enum: 5 / 10) - clip length in seconds. Longer costs more.
• aspect_ratio (enum: 16:9 / 9:16 / 1:1) - landscape, vertical, or square.
• mode (enum: standard / pro) - quality tier. pro is higher quality and pricier; standard is the cheaper default.
• fal_key (STRING) - your fal.ai API key. Required - no key, no generation.
• image (IMAGE, optional) - a starting frame for image-to-video. Leave it empty for pure text-to-video.

The output is a STRING - the URL of the generated video on fal's storage. To get the file into your graph, you'll typically download it from that URL with a load-from-URL step; the node hands you the link, not the pixels.

How to install it

Ships with the Mixlab pack.

• ComfyUI Manager: search comfyui-mixlab-nodes, install, restart.
• Manual: cd ComfyUI/custom_nodes && git clone https://github.com/shadowcz007/comfyui-mixlab-nodes.git, install requirements (install.bat on Windows or ../../../python_embeded/python.exe -s -m pip install -r requirements.txt), restart.

You'll also need a fal.ai account and an API key from your fal dashboard, and credits on that account.

Common issues

Auth errors / nothing generates: the usual culprit is the fal_key - missing, mistyped, or an account with no credits. This node can't work without a valid, funded fal key. Note the key is a secret; be careful not to share a workflow JSON with your key baked into it.

It's slow / seems to hang: hosted video generation takes real time on fal's side, and the node is waiting on that. That's expected, not a freeze - Kling jobs aren't instant.

I got a URL, not a video: by design. The output is the link. Add a download/load-from-URL node to pull the actual file into your graph if you want to process or save it locally.

Cost surprise: every run bills fal, and pro mode plus 10-second clips is the expensive corner. If you're iterating on prompts, do it in standard at 5 seconds to keep the meter down, then switch to pro for the final. This is the one node in the pack where hitting "queue" repeatedly actually costs you money.