Kling 2.6 · Text to Video

No image, just a prompt and a prayer. Kling 2.6 Text to Video starts around 10.35 credits per second - call it ten cents a second - and unlike the 2.5 Turbo node it gives you a generate_audio switch, so you can make a silent clip and not pay for a soundtrack you'll replace in an editor anyway.

That switch is the reason this node exists next to the cheaper one. If you want Kling motion on a budget with your own sound design on top, this is the node. If you want the cheapest possible Kling shot and you're fine with whatever audio comes attached, use 2.5 Turbo.

What happens when you queue

The node assembles a request from the widgets, asks PopNow for a quote, submits the task under an idempotency key, polls it, downloads the MP4, and returns it as a VIDEO. Kuaishou's model does the rendering; your machine does nothing but talk HTTP. There's no resolution widget here - you're not picking 720p versus 1080p on this node - and no image socket, so every take is invented from text.

Inputs that matter

prompt is the whole game. With audio on, put spoken lines in quotes and name the sounds you want, because the model is generating both streams together. Keep the picture description and the sound description in separate sentences; smearing them together is how you get a character who says "thunder rumbles" instead of a thunderclap.

aspect_ratio - 16:9, 9:16, or 1:1.

generate_audio - off by default, and audio costs more per second. The badge on the node updates when you flip it, so you can see the real rate before committing.

duration_seconds - 5 or 10. Per-second billing means the length dropdown is a price dropdown.

negative_prompt - a short list. guidance_scale - 0 to 1, default 0.5, adherence dial rather than a CFG value. Push it up only when the model keeps wandering off the prompt.

Outputs

video (VIDEO) and task id (STRING). The video is a normal ComfyUI value: Save Video, or another PopNow node if you're extending or editing it. The task id is your reference in the API console - worth copying into a text note when you're generating variants, because the console is where a cancelled or failed run's record lives.

Install

ComfyUI Manager → search PopNow → Install → restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/popnow-ai/popnow-comfyui

Then a key from popnow.ai/api/console/keys into Settings > PopNow (or POPNOW_API_KEY in the environment), and a second restart so the node list is built from your account. There are no model files and no extra Python packages - the pack declares no dependencies and only uses what ComfyUI already ships. Remove the old single-file custom_nodes/popnow_api.py if you installed that back when it was the only option.

Practical gotchas

Cost creep is the real hazard, not crashes. At ten cents a second, a session of "let me just try ten variations" is a five-second clip times ten, and then you do it again at ten seconds. Set a ceiling first: Settings > PopNow > Max credits per run. Any run quoted above it aborts before anything is held, which is exactly the guardrail you want while learning how a model reads your prompts.

Cancelling is not a refund. ComfyUI's interrupt stops the wait; the task keeps generating on PopNow, gets charged if it succeeds, and you'll find it in the console later.

Errors worth recognising: PopNow 401 is a bad or revoked key. 40201 is an empty wallet. 40004 means the settings combination isn't servable - change the field the message names. Anything ending in request_id ... is a real support ticket; that id pins the exact request. Failed generations are refunded, so a content_rejected costs you time, not credits.