Kling 2.6 · Text to Video

Kling 2.6 without a reference frame: you write the scene, pick a frame and a length, and it renders 1080p at $0.069 per second - or $0.138 if you flip sound on. Five seconds silent is 35 cents; five seconds with a generated soundtrack is 69 cents. There's no resolution choice on this node, so the only two dials that move the bill are duration and audio, which is honestly refreshing.

The appeal over the image-to-video sibling is speed of thought. No still to generate, no folder to dig through, just a sentence and a queue. The cost is control: text-to-video can hold a place, a mood and a camera move, but it cannot hold a specific person or product across takes, and it won't hit a composition you had in your head. Use it for establishing shots, backgrounds, title sequences, B-roll, and anything where the world is the subject. When a character has to look the same in take two, generate a still instead and animate that.

The inputs, in the order they matter

• prompt - scene, action, camera. With sound on, add dialogue in quotes and the sounds you want; the audio pass reads the same prompt, so a scene you describe as "rain hammering on a corrugated roof" actually comes back sounding like that.
• negative_prompt - what must not appear. A short list of concrete things works; quality-adjectives don't.
• aspect_ratio - 16:9 (default), 9:16, 1:1.
• generate_audio - off by default. On adds a native soundtrack of voices, effects and ambience and doubles the per-second rate. There's no partial credit: it's audio or silence.
• duration_seconds - 5 or 10, billed per second.
• guidance_scale - 0 to 1, default 0.5. Higher sticks closer to your prompt, lower gives the model latitude. It's a literalness dial, not CFG, and 0.5 is the middle for a reason.

Outputs: video (VIDEO - Save Video, a preview node, or a video-stitch node) and task id (STRING, the run's console handle for support or logs).

Install

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

Restart ComfyUI, then Settings > SpicyAPI → paste a key from spicyapi.ai/console/keys → restart once more so the node list is fetched from your account. Manager search is "SpicyAPI". Server installs: set SPICY_API_KEY and skip the panel. No GPU, no weights, no extra Python packages - Python 3.10+ and ComfyUI's V3 node API is the entire requirement.

Getting your money's worth

Write the camera into the prompt. The single biggest difference between a clip that looks generated and one that looks shot is a described camera move - "slow push in", "locked-off wide", "handheld pan left". Kling is good at this and it's free.

Keep it to five seconds unless you need ten. Identity, lighting and set details all drift with length, and the per-second meter keeps running.

Decide the audio question once, up front. If you're going to replace the track in an editor anyway, silent costs half as much and the picture is identical. If you're publishing straight out of ComfyUI, the audio pass is genuinely the reason to use a hosted model at all.

Set Max cost per run in the SpicyAPI settings before you start iterating. Every run is quoted first and anything above the ceiling aborts before charging, and the node prints the actual charge when the job completes. Remember that seeds default to fixed - queueing the same graph twice returns the first result, not a second bill - and flip control after generate to randomize when you want a new sample. Interrupting a run stops your wait but not the job: it carries on server-side, is charged if it succeeds, and its result is available in your console for up to 14 days.

Errors, briefly: 40310 is an unverified account email, 40004 means the option combination you picked isn't served, and a message mentioning a proxy or firewall means your HTTPS_PROXY isn't pointing anywhere useful. Add a new model to your account and you'll need a ComfyUI restart before it appears in the node search.