Kling v2.6 Pro Video Generation (fal)
Kling v2.6 Pro video, now with native audio
- image
- STRING
Kling is Kuaishou's closed video model, and it's been one of the strongest in the business for a while - the kind of motion quality and prompt-following that open models keep chasing. Kling26Pro runs the v2.6 Pro endpoint through fal, inside ComfyUI. Nothing runs locally; you send a prompt (and optionally a starting image), fal generates the clip on Kling's side, and you get a URL back. The headline for 2.6 is native audio - it can generate sound with the video, which until recently was the one thing closed models like Veo could do that nobody else could.
It's part of gokayfem's ComfyUI-fal-API pack. Reach for it when you want top-tier closed-model video quality and you're fine paying per clip.
How it works
Text-to-video or image-to-video, your choice: leave the image input empty for pure text-to-video, or plug an image in to animate from it. fal handles the generation remotely and returns a string URL to the result. The pattern across all these video nodes is the same - the output is a URL, not a native VIDEO, so save it before it expires.
The inputs and outputs that matter
prompt- describe the scene and the motion you want.duration-5or10seconds. Ten costs more; five is plenty for testing.image(optional) - supply one and you're doing image-to-video; omit it for text-to-video.cfg_scale(0–1, default 0.5) - Kling's guidance dial runs on a 0–1 scale, not the 1–15 you're used to from local samplers. Higher sticks closer to the prompt; 0.5 is a good middle.generate_audio(default on) - the 2.6 feature. Leave it on for synced sound, off if you only want the visuals.
There's also a negative_prompt (defaults to blur, distort, and low quality) and aspect_ratio (16:9 / 9:16 / 1:1). The output is a single STRING - the video URL.
How to install it
Via ComfyUI Manager: search ComfyUI-fal-API, install, restart. Manually:
cd ComfyUI/custom_nodes
git clone https://github.com/gokayfem/ComfyUI-fal-API.git
pip install -r ComfyUI-fal-API/requirements.txt
No models to download. Add a fal API key to config.ini under [API] (from config.ini.example) or export FAL_KEY=your_key, restart, and find it under FAL/VideoGeneration.
Common issues & troubleshooting
- Auth error.
FAL_KEYnot set. Fix and restart. cfg_scaleacting weird. Remember it's 0–1 here. If you typed7expecting sampler-style guidance, that's out of range - dial it back to something like 0.5.- The URL expired. fal result links don't last. Route the
STRINGoutput into the pack's Save-Media-from-URL node, or download it immediately. - A run fails with an error. The pack shows fal's real message - content policy, quota, or a bad parameter. Read it.
- Cost. Kling Pro is one of the pricier endpoints, and 10-second clips cost more than 5. The result cache serves repeat identical calls free (
force_rerunbypasses); the spend guard caps a session - turn it on before a long night of iterating. - Windows Portable. Missing
fal-client?python_embeded\python.exe -m pip install fal-client.
If you want the best closed-model video quality accessible from a ComfyUI graph - and especially if you want audio baked in - this is the node. Just keep an eye on the meter.
Inputs (7)
| Name | Type | Default | Description |
|---|---|---|---|
| prompt | STRING | — | |
| duration | COMBO | 5 | 2 options: 5, 10 |
| imageopt | IMAGE | — | |
| aspect_ratioopt | COMBO | 16:9 | 3 options: 16:9, 9:16, 1:1 |
| negative_promptopt | STRING | blur, distort, and low quality | — |
| cfg_scaleopt | FLOAT | 0.50–1 | — |
| generate_audioopt | BOOLEAN | true | — |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| STRING | STRING | — |