Kling 2.1 Master · Text to Video

No image, no anchor, full price

Text-to-video is the harder task and everyone knows it - with an opening frame the model only has to continue something, without one it has to invent the world, the subject and the camera at once. Kling 2.1 Master does it at a fixed 1080p, silently, at 32.4 credits per second. A five-second clip lands around 162 credits; ten seconds doubles it.

So why this over the image-to-video Master node? Speed of thought. You don't need a frame first, so the loop from "I wonder if…" to a moving picture is one node. If you already have a strong still - from Nano Banana Pro, from Qwen-Image, from your own local stack - the I2V variant gives you more control for the same money, because you're constraining the first frame instead of hoping for one.

How it works

Your prompt goes to PopNow, gets quoted, is submitted as a task under an idempotency key so a retried network call can't bill twice, and the node polls until the render settles. Then it downloads the MP4 and hands you a proper ComfyUI VIDEO. No model loads locally; the pack ships zero dependencies and rides on the libraries ComfyUI already has.

Because the whole thing is remote, nothing about your hardware matters - laptop, Mac, CPU-only box, all identical. What matters is the wallet and the network path: if a proxy is set in HTTPS_PROXY and it's broken, you'll get a message about interception rather than a mysterious hang.

Inputs

• prompt - the scene, the action, the camera. One action per clip. If you want a character to walk, turn, and then look up, you're describing a shot list, not a five-second clip.
• aspect_ratio - 16:9 by default. Set it before you queue; you're paying by the second and a reframed idea is a wasted render.
• duration_seconds - 5 or 10, billed per second.
• negative_prompt - short list, three or four items. Long negatives tend to come back as the thing you banned.
• guidance_scale - 0 to 1, default 0.5. Not CFG. Raise it toward 1 to make the model obey you literally; drop it to give it creative room. If you're used to turning CFG to 7 on a local video model, unlearn that here.

Outputs: video (VIDEO, wire it to Save Video) and task id (STRING).

No audio. No resolution widget either - Master is 1080p by definition. If you want a soundtrack generated with the clip, use Kling 2.5 Turbo, which is far cheaper per second and ships sound.

Install

Manager → search PopNow → Install, or:

comfy node install popnow-comfyui

Restart, then key: create one at popnow.ai/api/console/keys and paste it in Settings > PopNow, or export POPNOW_API_KEY on a server. Restart once more so the node catalog rebuilds from your account. Delete custom_nodes/popnow_api.py if the old single-file node is still installed. A recent ComfyUI is required - the pack uses the V3 node API.

Money and mistakes

Before anything else, set Settings > PopNow > Max credits per run. This is a per-second model at the top of the price table, and that ceiling blocks a run whose quote exceeds it before a single credit is held. Every run is quoted anyway, and the final charge appears on the node when it finishes - read it.

PopNow 401 is a key problem, 40201 is an empty wallet, 400 names a rejected value, 40004 means the combination isn't servable (aspect ratio and duration are the usual pair), 40310 is email verification. Declines and failures - content_rejected, generation_failed - are refunded.

Two things people learn the hard way. Text-to-video wanders: five seconds of a coherent scene beats ten seconds of a scene that melts, so start short. And cancelling doesn't cancel - the render continues on PopNow and is charged if it succeeds, with the result sitting in your console logs afterwards.

Last: this vendor is genuinely new, with no community trail to learn from - searching the usual forums for PopNow returns nothing. Treat the first dozen runs as a trial, keep the key in the environment so it never lands in a shared workflow, and don't build a client deliverable on a pack you haven't watched behave.