Kling Text to Video
Still the one to learn first
- VIDEO
- video_id
- duration
The plain Kling Text to Video node is the workhorse of the Kling family - no gimmicks, no storyboards, no audio, just prompt-to-video on Kling's current-ish model line. If you want to understand how all the other Kling nodes behave, start here. It's the node people reach for first because it's the least fiddly: one prompt, one mode dropdown, a video out.
Mechanically it's the classic partner-node flow. The node fires a text-to-video request at Kling's servers through your Comfy account, polls until the task finishes, and returns the finished clip. Nothing renders on your machine - no model files, no VRAM worries - and each generation bills to your Comfy account credits. The node even shows a price badge that updates from your chosen mode.
The inputs that matter:
- prompt / negative_prompt - your positive and negative descriptions. Kling responds well to explicit motion and camera language.
- cfg_scale - 0 to 1, default 1. Note the range: Kling's CFG is a 0–1 scale, not the 1–10 you're used to from Stable Diffusion. It's a guidance strength, and the default is fine for most work.
- aspect_ratio - 16:9, 9:16, or 1:1.
- mode - the sneaky one. It looks like one dropdown but it packs three settings in the format mode / duration / model. Currently it offers
pro mode / 5s duration / kling-v2-5-turboandpro mode / 10s duration / kling-v2-5-turbo. So "pick a mode" is really "pick a length and a model." The node only lists combos the current API actually supports, so what you see is what you get.
Outputs are VIDEO plus video_id and duration strings. The video_id is genuinely useful: Kling's ecosystem (video extend, lip-sync) can take a prior video's id, so wire it to a primitive if you plan to chain generations.
Where people get caught: the cfg range trips up SD refugees, and the mode dropdown confuses people who expect a separate duration slider - it's all in that one combo. Beyond that, remember the model you're on. The v2.5-turbo option is current as of this writing, but Kling retires legacy models on a schedule; if the dropdown ever looks thin, it's because retired options were pruned, not because you broke something. The bigger picture: Kling 3.0 (KlingVideoNode) has since landed with longer durations and multi-shot, so this node is the dependable mid-tier, not the top of the line.
House rules for every partner node: log into your Comfy account or it refuses to run with an auth error, and if the whole API-node family clutters your node search, --disable-api-nodes in your launch args removes them.
Inputs (5)
| Name | Type | Default | Description |
|---|---|---|---|
| prompt | STRING | Positive text prompt | |
| negative_prompt | STRING | Negative text prompt | |
| cfg_scale | FLOAT | 1.000–1 | — |
| aspect_ratio | COMBO | 16:9 | 3 options: 16:9, 9:16, 1:1 |
| mode | COMBO | pro mode / 5s duration / kling-v2-5-turbo | The configuration to use for the video generation following the format: mode / duration / model_name. |
Outputs (3)
| Name | Type | Description |
|---|---|---|
| VIDEO | VIDEO | — |
| video_id | STRING | — |
| duration | STRING | — |