Kling 2.6 · Image to Video

If you already have the still, this is the node most people should reach for. Kling 2.6 at 1080p, five or ten seconds, $0.069 per second silent - and the person, outfit and setting hold from your image while the prompt adds the action. Turn audio on and the rate doubles to $0.138/s. That toggle is the most consequential widget on the node, and it's worth being deliberate about it: a five-second clip is 35 cents silent and 69 cents with sound.

Compared with 2.5 Turbo, you're paying a little more per second and getting a newer model with a real audio choice rather than audio welded in. Compared with the 3.0 nodes, you're giving up resolution tiers, durations beyond ten seconds, and shot planning, and paying a good deal less for it. For a talking-head-ish or single-subject shot under ten seconds, that's a sensible place to sit.

The inputs

• image_url - the opening frame, IMAGE, JPEG or PNG, uploaded for the run (PNG under the 10 MiB ceiling, JPEG above). The clip keeps the image's aspect ratio.
• prompt - what happens after the frame: the action, the camera, and - with sound on - the sounds. Put dialogue in quotes if you want it spoken rather than narrated.
• negative_prompt - what must not appear. This generation still listens to it.
• generate_audio - a switch, off by default. On gives you a native soundtrack of voices, effects and ambience with the picture, and it costs more per second. Off is silent. This is the switch to check before you queue a batch.
• duration_seconds - 5 or 10, billed per second.
• guidance_scale - 0 to 1, default 0.5. Literalness dial, not CFG: higher follows the prompt more closely, lower gives the model more freedom.

Outputs: video (VIDEO - Save Video or a preview node; it's not an image tensor, and if you generated audio, the audio rides inside the video file) 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 rebuilt from your account's catalogue. Manager search: "SpicyAPI". Headless machine: SPICY_API_KEY in the environment. The pack installs no dependencies and downloads no models - the generation happens on SpicyAPI's servers, which is also why a laptop or a CPU-only box runs this exactly as well as a 4090 does.

What actually goes wrong

Sticker shock from the audio toggle. People switch it on once to hear what it does, leave it on, then wonder why a session of ten clips cost seven dollars instead of three-fifty. The node's own badge updates as you change options, so read it before queueing.

Ten-second clips drift. Identity holds well over five seconds and starts to slide over ten - faces soften, clothing details wander. If a shot needs length, consider two five-second clips cut together rather than one ten-second render.

Negative prompts are for objects, not vibes. A short list of things you don't want ("text, watermark, extra fingers, blur") does work; "bad quality, low res" does approximately nothing, same as everywhere else.

Cost mechanics you should set once: put a number in Max cost per run - every job is quoted first and aborted, uncharged, if it's above your ceiling - and know that seeds start on fixed so re-queueing the same graph reuses the first result rather than billing you for an identical clip. Switch control after generate to randomize when you want a fresh take. Interrupting a run doesn't stop the server-side job: it keeps running, gets charged if it succeeds, and its result hangs around in your console for up to 14 days.

Standard errors: No SpicyAPI key is set means the settings step didn't take; 40310 means verify your email; 40004 means the specific combination you chose isn't served - change the field the message names. A message about a proxy or firewall intercepting the request means HTTPS_PROXY is set but not working. Keep the task id output visible and you'll always have the identifier support needs.