Kling - V3 Turbo Image to Video

What it is

V3 Turbo is the current-speed tier of Kling's third generation: faster and cheaper than the Omni flagships, more capable than the V2.x line. This node takes your still and animates it - with a mandatory prompt, a 720p/1080p resolution choice, and any duration from 3 to 15 seconds.

KIE_Next_kling_v3_turbo_image_to_video_1caa8743, in ComfyUI-KIE-Nodes-Next, KIE Next / Video / Kling. Kuaishou's closed model, no downloadable weights, so the render is a paid call to KIE.ai on a KIE account - the third-party reseller path, not ComfyUI's official Partner Nodes.

Inputs

• image_urls (IMAGE, required) - the first frame. Despite the plural name (it's KIE's field name, and on other endpoints it accepts a batch) this socket takes your image happily; the pack uploads it for you as PNG.
• prompt - required, up to 2500 characters. On image-to-video the useful prompt describes motion, not the picture. The shipped default is a chatty two-character villa scene with dialogue; you don't need that. "She turns her head toward the window, curtain drifts, slow dolly in" is the register.
• duration - and here's the thing: it's a plain text field, not a dropdown. KIE documents 3s–15s, the default is 5, and you type whatever you want. That means you can send 7 and mean it, and you can also typo 7s or seven and find out the hard way. Treat it as a text box with a known-good vocabulary: digits.
• resolution - 720p or 1080p. No 4K here; that's the Omni tier.
• timeout_seconds - 1200 seconds of polling. callback_url - leave empty.

Note what's absent: no seed, no negative prompt, no cfg. The V3 Turbo endpoints are leaner than the V2.x ones. If you liked tuning cfg_scale on V2.1 Master, that dial doesn't exist on this node.

Mechanism

Upload the image, POST the job to KIE's createTask (kling/v3-turbo-image-to-video), poll recordInfo until the task reports success - interval growing from ~2.5s toward a 15s cap, with retries and backoff handled inside the client for 429s and transient server errors - then download the MP4 into output/KIE-Results/videos/ and return it as a native ComfyUI VIDEO. Downstream, it behaves like any clip: save it, extract frames, upscale it locally, chain it into an interpolation node.

Because a 15-second render can genuinely take a while, and because the poller gives up with a message telling you the state and progress it last saw, raise timeout_seconds rather than resubmitting. Resubmitting is a second bill.

Outputs

video, url, all_urls_json, task_id, raw_json, credits_consumed, credits_left. The credit receipt prints on the node after each successful run, which is the fastest way to learn what 1080p for 12 seconds costs before you plan a twelve-shot sequence. credits_left returning -1 means KIE's balance endpoint was unavailable for that check; the render is fine.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/felipederosilva/ComfyUI-KIE-Nodes-Next

Restart ComfyUI, then Settings → KIE.ai Nodes Next → Connection → Paste / replace KIE.ai API key. Validation happens before the key is stored, the frontend only sees a masked fingerprint, and you get a ✅ CONNECTED row plus your balance. Servers can set KIE_API_KEY instead; it takes priority. Dependencies are requests, Pillow, numpy and PyYAML, nothing more. Manager install works where the pack is listed, and the repo ships a ZIP plus a one-click Windows installer under dist/.

Traps

• A junk duration. It's free text. Digits only. If a job fails on an unhelpful provider error, check this field first.
• Prompting the image you already sent. Describe the change.
• Expecting cheap at 15 seconds. Duration scales the cost. Test at 5 seconds, then go long once you like the motion.
• Missing node on a fresh install. The pack generates model nodes from KIE's documentation index at startup and re-syncs at most once per day; the first launch shows a small bootstrap set. Restart and check again in a minute.
• Video-output errors about comfy_api.latest. Your ComfyUI build is too old to wrap the downloaded clip as a VIDEO. Update ComfyUI.