EvoLink Kling V3 Text-to-Video (Official)

Kling V3 is Kuaishou's closed video model - you cannot download it, full stop. So if you want a Kling clip in the middle of a ComfyUI workflow, your only door is an API, and this node is that door. It's text-to-video: describe a shot, get a video file back that wires straight into ComfyUI's native Save Video node. No local GPU, no model files, no ffmpeg wrangling. If you've ever stared at a Kling render on someone else's timeline and wished you had it, this is the low-friction version.

What it does

EvoLink Kling V3 Text-to-Video submits your prompt to EvoLink's cloud (POST /videos/generations), polls until the job finishes, downloads the result, and hands it to ComfyUI as a VIDEO object. Duration is 3 to 15 seconds, any integer, in 720p, 1080p, or 4K. It also supports sound and a negative prompt, which is more knobs than most video API nodes bother exposing.

The inputs you'll actually touch:

• prompt (required) - the shot description. Chinese works, up to 2500 characters. This is the whole job; Kling is one of the video models where prompt quality shows up immediately.
• duration - 3–15 seconds, default 5. Billed by duration, so a 5-second proof-of-concept costs less than a 15-second final.
• quality - 720p / 1080p / 4k. Start at 720p; Kling's 4K is where a session starts to hurt.
• sound - off by default. Leave it off while you're iterating and flip it on for the take you're keeping.
• negative_prompt - what you don't want in the frame. Handy for "no text, no watermark" cleanups.
• api_key and timeout_seconds - the key you paste once; the timeout defaults to a generous 1200 seconds (20 minutes), which video jobs genuinely need.

Outputs: video (already downloaded, connect it to Save Video), task_id (your receipt for the dashboard logs), status_info (a human-readable summary including credits spent), result_urls (raw links, 24-hour expiry), and response_json for downstream nodes that want to parse the raw API response.

Install and first run

Same story for every EvoLink node, so do it once and you're set. ComfyUI Manager → Custom Nodes Manager → search EvoLink → install, then fully restart ComfyUI (close the console, not just refresh the browser). Or the manual route:

cd ComfyUI/custom_nodes
git clone https://github.com/deeplearning-goethe/comfyui-evolink

The only dependency is requests, which ComfyUI already has. No models download - nothing runs locally, that's the point. Grab a key from evolink.ai/dashboard/keys (starts with sk-) and paste it into the api_key field once; it saves to a local config file and you can leave the box blank forever after. The key never ends up in a shared workflow JSON.

Gotchas

The big one: nothing shows up unless you connect the video output to a Save Video node. Re-run after connecting and you won't be double-charged - unchanged params reuse ComfyUI's cache, and failed or moderation-blocked tasks aren't billed. The VIDEO output type needs a reasonably recent ComfyUI (it relies on the comfy_api input path); if the node errors with something about that, upgrade ComfyUI first. result_urls die after 24 hours, so save locally if you want to keep it. And on the money side: this is a metered cloud call, the same tradeoff as every API node - per-render cost, your prompt and any reference images leave the machine. Fine for a clip you can't get any other way; a bad default for something you could render locally.