Kling 4.0 · Text to Video

Four widgets, a prompt box, and a per-second rate of from 32.4 credits - three times the Kling 2.5 Turbo node sitting a few entries above it in the same menu. Kling 4.0 Text to Video is what you use when the shot has to be right the first time and you're willing to pay for a 30-second 1080p take with sound baked in. It's not the node you learn on.

What it actually is

A client. Your prompt and settings become a JSON request; PopNow quotes a price, holds it, submits the task with an idempotency key so a retried HTTP call can't create two jobs, polls until the task finishes, then downloads the MP4 and hands it to your graph as a VIDEO. Kuaishou renders on their hardware. Nothing about your machine matters beyond an internet connection and ComfyUI starting up.

The node's description also flags that this tier "may queue when busy". Long Queued stretches are the model's schedule, not a bug in your graph.

Inputs

prompt - scene, action, camera and the sound, since audio isn't a separate switch here. Put spoken lines in quotes. This is a node where the prompt is doing two jobs, so give the sound its own sentence: "Rain hammers the awning. She looks up and says 'not again'." sounds like nothing, but it renders like something.

aspect_ratio - 16:9, 9:16, 1:1, or 21:9. The ultrawide option is one of the reasons to be on 4.0 at all; it's absent from the 3.0 node.

resolution - 720p or 1080p. 1080p costs more per second.

duration_seconds - whole seconds from 3 to 30, billed per second. Thirty seconds at base rate is 972 credits, just under a dollar, for one take.

No guidance dial, no negative prompt box, no shot list, no keyframes - those live on the image-to-video and reference siblings. This is the purest "prompt in, clip out" node in the family.

Outputs

video (VIDEO) and task id (STRING). Video to Save Video or into a PopNow edit/extend node. The task id is worth keeping when you're comparing takes, since it's the lookup key in the API console.

Install

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

Restart, then paste a key from popnow.ai/api/console/keys into Settings > PopNow (or set POPNOW_API_KEY), restart once more so your account's catalog becomes the node list. ComfyUI Manager has it under PopNow too. Nothing else to install: no weights, no requirements. Delete custom_nodes/popnow_api.py if you had the older single-file node.

The money talk

Set Settings > PopNow > Max credits per run before you touch this node. It's the difference between "I wonder what 30 seconds looks like" costing a dollar and costing nothing. Any run quoted above the ceiling aborts before credits are held.

Iterate cheap: test the idea as a 5-second 720p clip at a quarter the cost, then spend on the version you're keeping. The prompt is the variable that matters most at this tier, and prompt work is free at 2.5 Turbo rates.

Cancelling in ComfyUI doesn't cancel the render. It stops you waiting; the job keeps running on PopNow, is charged if it succeeds, and the output waits in the API console. Check there before re-queueing something you already paid for.

Codes worth memorising: 40201 (no credits), 401 (bad key), 40310 (email not verified), 40004 (a settings combination the model won't serve - usually length), content_rejected / generation_failed (refunded). Anything ending in request_id ... is a support ticket with the exact request attached.

Last thing: newer isn't automatically better for your footage. Kling's generations have split opinion before, and the honest move is to render the same prompt on 2.5 Turbo, 3.0 and 4.0 once, then decide with your own eyes which one you'd pay for.