视频｜Kling 图生视频

Kling is one of the Chinese video labs the community actually knows by name - when people ask "what made this?" in video threads, Kling is a recurring answer alongside Veo and Hailuo. This node gives you Kling's image-to-video inside ComfyUI through the Tikpan reseller: a first-frame image in, an animated clip out, with an optional tail frame to bookend the motion. It defaults to kling-v2-5-turbo, which the tooltip sells as the value pick - faster and cheaper, with the heavier v2.6/v3 models a dropdown away.

As with every node in this pack, it's a metered cloud call: your sk- key from https://tikpan.com, pay per task, images and prompts leave your machine. Kling's closed models are exactly the "no open weights, so local isn't an option" category the KB's API-node doc describes - there is no Kling you can run at home, which is the entire reason this node exists.

How it works

The required 首帧图 is your first frame - the image the video starts from. An optional 尾帧图 sends image_tail, letting you specify where the motion should end; that's the difference between "anything happens" and "this video ends here." The node submits to Kling's image2video endpoint, polls, downloads, and returns the standard five outputs (🎬_视频输出, 📁_本地保存路径, 🆔_任务ID, 🔗_视频云端直链, 📄_完整日志).

Beyond the frame ports, the interesting knobs are the Kling-native params most packs strip out:

• 模式 - std (faster/cheaper) vs pro (higher quality).
• 负向提示词 - what you don't want in the frame; leave empty to omit.
• CFG Scale - defaults to 0.5, Kling's cfg_scale passthrough.
• 生成声音 - sends sound=on, but only v2.6 and up support it, and it adds time and cost.
• 相机控制JSON - raw Kling camera_control JSON, for the crowd who wants literal camera moves.
• 视频时长 - 5s or 10s; 画面比例 - 16:9 / 9:16 / 1:1.

Install and first run

cd ComfyUI/custom_nodes
git clone https://github.com/htrert/ComfyUI-Tikpan-Pro

Restart (or Manager → "Tikpan"), paste your key, load a first frame, set std + 5 seconds + 480p-equivalent, run. Nothing to download; the model is server-side. Once the loop works, experiment with 尾帧图 - it's the cheapest way to add intent to a clip.

Common issues

The pack-wide trio (401 key, 402 balance, 429 rate) plus the standard video-task failures - timeouts (hold the 任务ID, recover via the task-query node rather than paying again), and charges with no file (read 完整日志 to see if the render actually finished). Two Kling-specific gotchas: 生成声音 silently does nothing on v2.5 Turbo (it needs v2.6+ - the tooltip says so, easy to miss), and a first frame with a big file can stall the upload, so compress it or use a hosted URL. If the VIDEO output won't preview, check 本地保存路径 first - the clip is usually on disk.