Vidu Q4 Image To Video

Most image-to-video models give you four to eight seconds and leave audio as somebody else's problem. Vidu Q4 does neither: this node takes a single image, a duration anywhere from three to sixteen seconds, and returns a clip with synchronized dialogue and sound effects generated in the same pass. If your shot needs a character to say something on a smooth camera move, that length-plus-audio combination is a genuinely different offer from what the Veo nodes in this pack give you.

It's also the simplest video node here. One image socket, one prompt field, two creative dials. No generation modes to get wrong.

How it works

A KIE Market job: the node uploads your connected image to KIE's file host (PNG-encoded), sends it as the image_url with your prompt, duration, resolution and audio flag, then polls the task until it completes and downloads the result into a ComfyUI VIDEO. The still is the first frame - the model is animating your picture, not re-interpreting it, so the better your source image, the better the clip. A local render or a hosted Nano Banana output both work as the start frame; that handoff is the whole reason these API nodes sit next to local ones in the same graph.

Set audio to false when you're going to lay your own track over the clip, which is what you'd do if you already have dialogue from a TTS pass and don't want the model inventing a competing one.

The inputs

• image_url - required, one image. Singular: this node wants a single frame, and if you connect a batch the pack's preflight will tell you so ("expects one image; received a batch of N. Select one frame first") rather than letting the provider fail.
• prompt - the motion instruction. The API documents it as optional (up to 20,000 characters), but the node lists it in the required group, so put something in it. Describe what moves and how the camera behaves: "the astronaut waves, camera slowly pushes in".
• duration - integer, 3 to 16 seconds, default 5. This is the headline feature and also the cost. Sixteen seconds is a lot of generation; start at five.
• resolution - 720p (default), 540p, 1080p, 2k, 4k. 540p is the cheap draft, 4K is the "I hope this is the take" button.
• audio - true by default: synchronized dialogue and sound effects. false for a silent clip.
• timeout_seconds - 1200s default. Long durations and 4K renders queue longer than a five-second 720p job; raise it rather than resubmitting.
• callback_url - empty for ComfyUI.

Outputs: video, url, all_urls_json, task_id, raw_json, credits_consumed, credits_left. Wire video into KIE • Preview Video or KIE • Save Video - the node isn't an output node, so an unconnected output means it never runs.

Installing

cd ComfyUI/custom_nodes
git clone https://github.com/felipederosilva/ComfyUI-KIE-Nodes-Next
cd ComfyUI-KIE-Nodes-Next && pip install -r requirements.txt
# restart ComfyUI

Manager search: "KIE.ai Nodes Next", or comfy node install kie-ai-nodes-next. Dependencies are requests, Pillow, numpy, PyYAML - no model weights, no VRAM, nothing video-related installed locally. Then paste your KIE key once in Settings → KIE.ai Nodes Next → Connection; it's verified against KIE before it replaces a saved key, stored in ComfyUI's user config area, and kept out of workflow JSON.

Where people get burned

Feeding it a batch is the first one, and it's a five-second fix: run the batch through a selector or feed one frame.

The second is forgetting what you're paying for. Every execution is a paid job, and duration × resolution is the price multiplier - a 16-second 4K render with audio is not the thing to discover a bad prompt on. Draft at 540p or 720p for five seconds, confirm the motion reads, then re-run with the settings you actually want. The right-click KIE Next · Check inputs & model controls menu shows the documented limits without uploading media or submitting anything, which is the free way to check a hunch.

Third, the structural one: the image you start from leaves your machine, gets forwarded to Vidu, and passes their content policy. Vidu is a closed model with no local equivalent at this length, so an API call is the honest price - but if privacy or filtering is the blocker, no node in this pack can route around it. Also worth knowing that the returned VIDEO needs a reasonably current ComfyUI; older builds raise an explicit "update ComfyUI" error instead of quietly handing you a broken object.