V1 Pro Fast Image to Video

This is the most stripped-down video node in the pack, and that's the feature. Give it an image and a prompt, pick 720p or 1080p, pick five or ten seconds, and go. No seed. No camera lock. No reference sockets, no aspect ratio dropdown, no first-and-last-frame contract - the image sets the aspect and the framing, the prompt sets the motion, and there's nothing else to misconfigure.

Compare it to its sibling V1 Pro Image to Video, which adds 480p, camera_fixed, seed and enable_safety_checker. If you're the kind of person who spends twenty minutes deciding what to put in a seed field, Fast will save you from yourself. If you need the camera locked off for a product shot, it won't, and you want the other node.

Inputs and outputs

Two required. image_url is a native IMAGE socket - connect anything your graph produces, and the pack uploads it to KIE and injects the URL so you never handle a URL by hand. prompt is multiline and long (docs allow up to 10,000 characters), and the shipped default is a good template for how to write it: a single continuous action, described in stages, with the camera implied - an espresso pouring into a cup, filling in visible steps. That "describe the sequence, not the end state" habit is what separates a clip that does something from a clip that just breathes.

Optional, and this really is the whole list: resolution (720p or 1080p - note there's no 480p here, so there's no ultra-cheap draft tier), duration (5 or 10), nsfw_checker (default true on this node), timeout_seconds, and callback_url - which you leave empty for normal ComfyUI use since it's for servers that can receive KIE's webhook.

Outputs: video (native ComfyUI VIDEO), url, all_urls_json, task_id, raw_json, credits_consumed, credits_left. Wire video into a save or preview node; the credit outputs plus the on-node receipt after each run are your spend tracking.

The tradeoff you're accepting

"Fast" here means fewer options, and fewer options means the model gets to make decisions you can't override. There's no aspect_ratio because the image dictates it; there's no camera_fixed because camera behaviour is the model's call; there's no seed because reproducibility wasn't the point. That's a reasonable design for this tier - it's the "I have a good still, animate it" node - but it does mean that when a clip keeps drifting or keeps pushing in, you have no switch to throw. Your only levers are the prompt and the input frame, and the input frame is the one that actually works.

So the workflow is: get the still right first, write the action as a sequence, and accept that a five-second clip from a single frame is a sketch rather than a precisely directed shot. If you need directorial control, this pack's V1 Pro or Seedance nodes have the dials for it, and they cost more.

Install

Manager → search KIE.ai Nodes Next (publisher id felipeosilva) → install. If Manager returns nothing, the README's registry publishing is still described as prepared rather than live, so clone it:

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

Restart ComfyUI afterwards. Dependencies are requests, Pillow, numpy, PyYAML and there are no model files to fetch - the model runs on KIE's infrastructure, which is why the requirements list is four libraries instead of a filesystem full of safetensors. Windows users can run the one-click .vbs installer in dist/; there's a full uninstaller too. Then set your key once in Settings → KIE.ai Nodes Next → Connection → Paste / replace KIE.ai API key. Validation happens before the key replaces a saved one, the UI shows only a masked fingerprint, and KIE_API_KEY in the environment takes priority on headless installs.

Gotchas

The node waits, so your queue does too. timeout_seconds defaults to 1200 with a ceiling of 7200, and a 10-second 1080p render is the case that needs it. A timed-out submission still costs credits, so set the timeout generously rather than re-queuing twice.

nsfw_checker defaults true here, unlike most of the V1 Lite nodes. If your output is coming back as an error rather than a clip, that's the first switch to look at - though ByteDance's model applies its own policy regardless, because it's a closed endpoint with no local weights to touch.

Nothing here is local, and that's the whole proposition - a per-call meter and your image leaving the machine, in exchange for a video model that will never run on your GPU. Judge it on that trade and not on the convenience.