Kling 2.6 Text-to-Video
The budget Kling tier with optional audio
- client
- video_url
Kling 2.6 Text-to-Video is the previous-generation Kling text-to-video node, and that's its real selling point: it's cheaper and less demanding than the 3.0 line while still producing solid, prompt-faithful clips. If you're doing volume - storyboards, drafts, shot ideas you'll never use - starting here instead of on Kling 3.0 keeps the bill down. When a clip actually matters, re-run it on 3.0 or Pro.
The model split is simple: Kling 2.6 (the standard tier) and Kling 2.6 Pro. Pro is the one with the interesting extras - cfg_scale for prompt adherence and sound for joint audio-video co-generation. Sound defaults to on for Pro, and the tooltip is honest that it doubles the cost. That's the biggest footgun in this node: if you don't want audio, turn it off or you're paying twice for every clip.
How it works
Same pattern as the rest of the pack's video nodes: you give it a prompt, it submits the job to WaveSpeed, polls until the clip is ready, and returns a video_url STRING. Feed that into Preview Anything to watch or download. There's an aspect_ratio dropdown (16:9, 9:16, 1:1) and duration is a fixed choice of 5 or 10 seconds - not a free range.
The inputs that matter
- prompt - describe scene, motion, and audio (Pro can generate sound, so prompt it like a filmmaker: "waves crashing, distant seagulls").
- model - Kling 2.6 or Kling 2.6 Pro.
- negative_prompt - what to keep out of visuals and audio.
- aspect_ratio - 16:9, 9:16, or 1:1.
- duration - 5 or 10 seconds.
- cfg_scale - Pro only, 0–1; higher follows the prompt more closely.
- sound - Pro only, default on, doubles cost.
The seed widget deserves attention: the tooltip is explicit that this endpoint has no seed parameter. It's cache control only - randomize re-runs a fresh, re-billed generation each queue pass; a fixed value lets ComfyUI reuse the cached video without calling the API.
Installing it
Part of the ERPK Collection (eRepublik-Labs/comfyui-nodes-erpk). ComfyUI Manager: search erpk → ERPK Custom Nodes → restart. Or:
cd ComfyUI/custom_nodes
git clone https://github.com/eRepublik-Labs/comfyui-nodes-erpk.git erpk
cd erpk
pip install -r requirements.txt
WaveSpeed API key required (Settings → ERPK → API Keys). No model downloads.
Common issues
- Audio you didn't ask for - Pro defaults
soundto on. Turn it off unless you want it; it doubles the per-clip price. - cfg_scale does nothing on standard - it's Pro-only. If you're on Kling 2.6, that input is inert.
- Duration surprises - it's a 5-or-10 dropdown, not a slider. There's no 7-second clip here.
- Seed isn't a seed - don't expect repeatability from a fixed seed; the API has no seed param. A fixed value only helps ComfyUI's cache.
Inputs (9)
| Name | Type | Default | Description |
|---|---|---|---|
| model | COMBO | Kling 2.6 | Model variant: Kling 2.6 (Std) or Kling 2.6 Pro (cfg_scale, sound) |
| prompt | STRING | Text description of the desired scene, motion, and audio | |
| clientopt | WAVESPEED_AI_API_CLIENT | WaveSpeed API client (optional if API key is configured in Settings) | |
| negative_promptopt | STRING | Elements to exclude from visuals and audio | |
| aspect_ratioopt | COMBO | 16:9 | Aspect ratio of the output video |
| durationopt | COMBO | 5 | Video duration in seconds (5 or 10) |
| cfg_scaleopt | FLOAT | 0.500–1 | Pro only: guidance strength (0.0-1.0); higher follows the prompt more closely |
| soundopt | BOOLEAN | true | Pro only: enable joint audio-video generation (doubles cost) |
| seedopt | INT | -1-1–2147483647 | Cache control: randomize re-runs generation each queue; a fixed value reuses the cached video. This endpoint has no seed parameter, so it is not sent to the API. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| video_url | STRING | — |