Kling 3.0 Turbo · Text to Video

Four widgets. That's the whole node: prompt, aspect ratio, resolution, duration. No negative prompt, no guidance scale, no shot planning, no audio switch - sound is generated with every clip and there's no way to turn it off. Turbo is the "stop configuring and just give me a clip" tier, and at $0.112 per second for 720p and $0.14 for 1080p you're paying a premium over the standard 3.0 node for that simplicity rather than saving money.

It's worth being clear about the trade, because "Turbo" reads like "cheaper" to most people. At 720p the standard Kling 3.0 node with audio switched on is $0.1008/s; Turbo is $0.112/s. At 1080p it's $0.1344 against $0.14. You're buying render speed and a smaller decision surface, and giving up negative prompts, guidance, shot lists and closing-frame control. For a "I need forty B-roll clips by Friday" job that's a good deal. For a shot you're going to fuss over, it isn't.

The four widgets

• prompt - scene, action, camera. Since audio always comes with the clip, write the sound you want as well: dialogue in quotes, ambience described plainly. The model voices and scores from the same prompt.
• aspect_ratio - 16:9 (default), 9:16 or 1:1. Vertical for shorts, obviously, and it's free to choose.
• resolution - 720p or 1080p. The node's badge updates with the rate.
• duration_seconds - whole seconds from 3 to 15, default 5. Billed per second, so this is the money dial: five seconds at 1080p is 70 cents, fifteen is $2.10.

Outputs are video (a VIDEO with the generated audio inside the file - Save Video or a preview node, not an image node) and task id (STRING, the exact handle for the run in the SpicyAPI console).

Install

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

Restart ComfyUI, open Settings > SpicyAPI, paste a key from spicyapi.ai/console/keys, restart once more so the node list is built from your account. Manager search: "SpicyAPI". Remote server: put SPICY_API_KEY in the environment instead. Nothing to download, nothing to compile, no GPU needed - the node posts your prompt to SpicyAPI and downloads the finished clip.

Without a negative prompt, prompt discipline is the tool

The absence of a negative prompt field isn't a disaster, it just moves the work into your wording. A few things that consistently help on hosted text-to-video models:

• Say the camera move. "Locked-off wide", "slow push in", "handheld tracking" - the model responds to these far more reliably than to mood adjectives, and a described camera is what separates a clip that reads as footage from one that reads as a render.
• Keep subjects singular and simple. One person, one action, one location. Text-to-video can't hold a specific face or product across takes, so don't ask it to.
• Ask for the sound explicitly if you care, because you're paying for it either way. "Rain on a tin roof, no music" is more useful than silence by omission.
• Three to five seconds is where drift stays invisible. At fifteen the model is inventing continuity for a long time, which is the general weakness of every video model rather than a Turbo quirk.

Money and mechanics

Set Max cost per run in the SpicyAPI settings before your first batch. Every job is quoted before it's submitted, anything above your ceiling aborts without charging, and the node prints the real charge when it lands. This is the node where a ceiling pays for itself, because fifteen-second renders at 1080p are $2.10 a pop and there's no resolution-shaped friction between you and the Run button.

Cancelling doesn't cancel. Interrupting a run stops your wait; the task keeps going on SpicyAPI, is charged if it succeeds, and its result is available in your console for up to 14 days. Seeds start on fixed so re-queueing the same graph returns the first result rather than a second bill - flip control after generate to randomize for a genuinely new sample.

If it doesn't run: No SpicyAPI key is set means the key never reached Settings > SpicyAPI (or the environment), 40310 means your account email isn't verified yet, 40004 means the specific combination of options isn't served - change the field named in the message. And the model list is fetched when ComfyUI starts, so anything new on your account shows up after a restart.