Kling 2.5 Turbo · Image to Video

The Kling node most people should pick

Native audio is the quiet revolution in video generation - Veo 3 was the model that made it obvious, and every serious release since has had to answer for it. Kling 2.5 Turbo ships sound with every take: ambient soundscape, foley, and the general sense that the clip happened somewhere rather than in a vacuum. On top of that it's 1080p and 8.4 credits per second, which is under a third of what the Master nodes charge for a silent clip.

If you're choosing between Kling nodes and don't have a specific reason to pick otherwise, pick this one. The image goes in, a five-second 1080p clip with audio comes out for about 42 credits, and the "does it have sound?" question never comes up again.

The mechanism

Your opening frame is encoded to PNG (or a high-quality JPEG above the 10 MiB upload ceiling), uploaded and cached for twelve hours on the content hash - so iterating on the prompt doesn't re-upload the same picture each time. Then: quote, submit under an idempotency key, poll, download, decode to VIDEO. There is no local model, no VRAM, no dependency install; the pack declares none and uses what ComfyUI already ships.

The audio arrives inside the video file rather than as a separate AUDIO output. That's worth knowing before you plan a graph around it: you can't feed the soundtrack into an audio node from here - you'd grab it downstream from the saved file.

Inputs

• prompt - what happens after the opening frame: the action, the camera. Since there's sound now, motion that implies noise (a door, rain, footsteps) is doing double duty - but the tooltip is about the visual action, and the model infers the rest.
• image_url - the opening frame, JPEG or PNG.
• duration_seconds - 5 or 10, billed per second. Turbo's job is iteration, so five seconds is usually the right answer until a take earns the extra length.
• negative_prompt - a short list, three or four things.
• guidance_scale - 0 to 1, default 0.5. This is not a 1-to-20 CFG knob; typing 7 does nothing useful. Higher is more literal, lower is more inventive.
• last_image_url (optional) - a closing frame the clip should end on. The cleanest way to get a controlled move rather than a drift, and the reason this node can do transitions the text-to-video variants can't.

Outputs: video (VIDEO, → Save Video) and task id (STRING).

No resolution widget: Turbo is 1080p, fixed. The only size dial you have is the duration.

Install

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

Restart, then create a key at popnow.ai/api/console/keys and paste it under Settings > PopNow - or set POPNOW_API_KEY if this is a headless server. Restart again so the node list rebuilds from your account's catalog. If you ever installed the old single-file popnow_api.py, delete it; it's the same node type.

Recent ComfyUI is required (V3 node API), and that's the whole prerequisite. Nothing downloads, no GPU involved.

Practical notes

Native audio is a coin flip in the current generation - sometimes it's a real soundscape, sometimes it's generic whoosh. You can't turn it off from this node, so if you need a specific soundtrack, plan on replacing the audio in your editor rather than fighting the model.

Codes: PopNow 401 key, 40201 credits, 400 rejected field, 40004 unservable combination, 40310 email verification, content_rejected / generation_failed refunded. Errors that end with a request_id are support tickets; copy the id.

Cancel doesn't cancel - the render continues and bills if it lands. And there's no seed on this node, so every queue is a new take and a new charge; get the frame and the sentence right before you start rolling.

As for the pack itself: PopNow is new, MIT-licensed and essentially undiscussed - there are no threads, no guides, no crowd-sourced list of gotchas. That means the price is attractive and the track record doesn't exist yet. Set Max credits per run in the PopNow settings, keep the key in the environment, and buy a small credit pack first.