KIE Kling 2.6 (T2V) [SL]

The text-to-video member of the Kling 2.6 family. No image input at all - you type a prompt, pick an aspect ratio and a duration, and Kling invents the whole clip from scratch. It's the node you use for the shot that doesn't exist yet, the establishing wide you can't source a still for, the "just show me something that looks like this" brainstorm pass.

Where the I2V sibling is about animating your subject, T2V is about generating the scene itself. If your workflow is "LLM writes a shot list, T2V generates the shots, I keep the good ones," this is the node in the middle of that loop. And unlike the I2V nodes, it defaults to portrait (9:16 aspect ratio) - a tell that Kling's T2V is aimed straight at vertical short-form content.

The inputs that matter

• prompt - everything, up to 2500 characters. There's no image to anchor it, so describe scene, subject, motion, and camera all in here.
• aspect_ratio - 1:1, 16:9, or 9:16 (default 9:16).
• duration - 5 or 10 seconds.
• sound - default off. Kling 2.6 can generate audio for a T2V clip the same way it does for I2V; flip it on if you want a complete draft rather than silent footage.

Output is a single video in ComfyUI's VIDEO format. Wire it into SaveVideo to keep it.

How it works

No uploads at all - this is the cheapest and fastest Kling in the pack because there's nothing to push up. The node creates a job against kling-2.6/text-to-video, polls recordInfo until the clip renders, downloads the MP4, and returns it as a VIDEO object. Same async machinery as the rest of the pack, same built-in retry on transient 429/5xx errors, same 1000-second timeout floor for the slow stuff.

Install

The whole pack installs one way:

cd ComfyUI/custom_nodes
git clone https://github.com/v3devv/ComfyUI-Kie-API-Serverless

or ComfyUI Manager → search "Kie-API-Serverless" → install → restart. Nothing to download, nothing to configure beyond pasting your KIE key into the node field.

Where people get burned

T2V gives you the least control of any node in the family - no reference image means character consistency across two separate T2V clips is a dice roll. If you need the same character in multiple shots, generate one keyframe and use an I2V node instead. The 10-second duration is where quality and cost both climb; 5 seconds is the sane default for drafts. And remember every clip is a metered call - this is a video model, so a single batch of "let's try a few variations" is a real line on your KIE balance.