视频｜Kling 文生视频

Kling's text-to-video, wrapped for ComfyUI: prompt in, cinematic clip out, with the same model ladder as the pack's Kling I2V node - kling-v2-5-turbo by default (the value tier), v2.6 and v3 a dropdown away. If you've used the image2video sibling, this is the exact same node minus the frame inputs: no first frame, no tail frame, just 生成指令 and the Kling-native knobs. Pure T2V is Kling's weakest mode relative to its competition, honestly - Kling shines when it has a first frame to work from - but for quick ad b-roll and concept clips where you have no image to start from, it's the right tool.

Same ground rules as every node in this pack: metered cloud API, sk- key from https://tikpan.com, your prompt leaves the machine, Kling's moderation applies. No local weights exist, so there's no local alternative for this specific model.

How it works

The 生成指令 is your whole creative input, so it carries the load - subject, action, camera, style, and mood all belong in there. The node submits to Kling's text2video endpoint, polls the task, downloads the result, and returns the standard five outputs: 🎬_视频输出 (VIDEO), 📁_本地保存路径, 🆔_任务ID, 🔗_视频云端直链, 📄_完整日志.

The knobs mirror the I2V node:

• 模式 - std (faster, cheaper) vs pro (higher quality).
• 负向提示词 - what to avoid; empty omits it.
• CFG Scale - Kling cfg_scale, default 0.5.
• 生成声音 - sound=on, but only v2.6+ models support it; it adds cost and wait.
• 相机控制JSON - raw Kling camera_control for deliberate 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, write a prompt, set std + 5s and run. Nothing downloads - the model lives on Kling's servers. Test the 相机控制JSON on a short clip if you want controlled camera moves; it's the feature that separates Kling from cheaper T2V options.

Common issues

The same pack-wide failure family: 401 bad key, 402 empty balance, 429 rate-limited, video task timeouts (recover with 任务ID via the task-query node instead of re-billing), and charges with no file (check 完整日志). One T2V-specific note: with no image anchor, prompt quality is everything - a thin prompt gives you a generic clip, and the "sound" toggle silently no-ops on the default v2.5 Turbo model, so switch to v2.6+ before you expect audio. And if the VIDEO output won't preview but the local path has a file, that's your ComfyUI missing a video-preview extension, not a failed render.