Generate AI Video

Runway's video line, called from inside ComfyUI. The company that put the SD 1.5 weights on HuggingFace in 2022 - and then deleted its org in 2024 - has been a closed commercial video and world-model shop ever since, with a Gen-series line and no open weights anywhere. This node routes their text-to-video and image-to-video endpoint through KIE, and it's the inexpensive way to have a closed video model sit between your local stages.

Where it fits

The canonical hybrid the community settled on is closed model for the frame, open model for the motion - generate a start frame somewhere strong, animate it locally with Wan or LTX. This node reverses the direction: it animates a frame for you, which is useful when the motion itself is the hard part, or when you're producing fast variants and don't want to babysit a local render.

Inputs

prompt (required) - subject, action, style, setting. Documented maximum is 1800 characters, and the tooltip's advice is the right advice: be specific. When you also connect an image, the prompt describes how that image should move or change.

duration (required) - the documented values are 5 and 10. The widget is a float that will cheerfully accept other numbers, so trust the docs, not the widget bounds.

quality (required) - a string, 720p or 1080p. Here's the constraint the tooltip spells out and the one people trip over: 1080p and a 10-second clip are mutually exclusive. Pick which one you want.

imageUrl - a native IMAGE socket. Connect a frame for image-to-video.

aspectRatio - 9:16, 16:9, 4:3, 1:1, 3:4. Required for text-to-video, and explicitly invalid when you pass an image, because the image decides the frame. Don't tune it and wonder why nothing changed.

waterMark - a string defaulting to kie.ai. An empty string means no watermark; anything else prints that text in the bottom-right corner. If you're delivering anything, set this deliberately.

timeout_seconds - 1200 default, 30–7200. callback_url - leave empty for normal ComfyUI use.

Outputs

video - a real ComfyUI VIDEO object, ready for Save Video or a preview node. url - the primary hosted result. all_urls_json - every URL the API reported; keep it, because video jobs sometimes return extra artifacts. task_id - the job handle. raw_json - the full response. credits_consumed and credits_left - floats, mirrored by the credits receipt the node draws on itself after a successful run.

Install

ComfyUI Manager (search the pack title), or:

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

Restart. Dependencies are requests, Pillow, numpy, PyYAML. No models, no local rendering, no VRAM - the node is a client and the meter runs per call.

Key: Settings → KIE.ai Nodes Next → Connection → Paste / replace KIE.ai API key. Validated against KIE before it replaces a working key; you'll see a masked fingerprint and CONNECTED when it takes. Headless installs can set KIE_API_KEY, which takes priority over the saved key.

Gotchas

The 1080p/10s conflict is a hard constraint, not a suggestion. Firing a job with both set is how you get a provider-side rejection after waiting.

Duration is a float in the UI. 5.5 is not a documented value. Type 5 or 10.

aspectRatio silently does nothing with an image attached. The docs say it's overridden by the image; if your output is the wrong shape, fix the source image, not the dropdown.

Watermark by default. kie.ai is in the field out of the box. It's easy to leave in place and notice three renders later.

Cards follow the provider. This pack generates its nodes from KIE's API documentation at startup and re-checks roughly daily, so widget changes arrive on a ComfyUI restart, and model availability on a reseller can shift faster than that.