Extend Veo3.1 Video · fast

Veo clips come in eight-second slices, which is fine until the shot needs to breathe. The Extend node is how you buy more time: it takes the task ID of a video you already generated and continues it, with a prompt describing what happens next. This is the fast tier of that operation - the middle price, the one you'll use most.

The whole thing only makes sense as a pair, so understand the chain before you wire it: Generate Veo3.1 Video hands you a task_id output. Extend Veo3.1 Video · fast takes that ID, adds a continuation prompt, and returns a new clip that picks up where the last one stopped. You can extend repeatedly - which is how people get 30-plus-second Veo sequences out of a model that renders eight at a time - and each extension is its own paid job.

How it works

Under the hood it's a KIE Market call: the node submits {"task_id": ..., "prompt": ..., "model": "fast"} shaped for the extend endpoint and polls the job until it reports success, then downloads the result and wraps it as a native ComfyUI VIDEO. The tier is baked into the node (fast), which is why there are three Extend nodes in the Veo3.1 folder rather than one with a dropdown. KIE bills each extension at the tier you picked.

The continuation is a new generation seeded from the previous clip, not a trim-and-append of stored frames. Continuity is good, not perfect - long chains drift, and the drift compounds. Two extensions usually look seamless; six usually don't.

The inputs

• task_id - required, and it must be a task ID the video generation API actually returned. It defaults to a placeholder string from the docs; wire the task_id output from your Generate node here instead of typing anything.
• prompt - what happens next. This is a motion prompt, not a re-description: "the dog keeps running, hurdles a low fence, camera tracks left". Re-describe the whole scene and you'll often get a soft reset of the framing.
• seeds - note the plural. Range 10000–99999, default 12345. Same seed plus same prompt trends toward similar motion; it's a nudge, not a lock.
• watermark - and this is the trap. It defaults to the literal string MyBrand, and if it's non-empty you get it burned into your video. Clear the field unless you actually want a watermark on your clip.
• timeout_seconds - 1200 by default; the extend call queues like any other render, so raise it before you assume it hung.
• callback_url - leave empty. It's for server-side workflows, not ComfyUI.

Outputs: video, url, all_urls_json, task_id (the new one - chain this into the next extension), raw_json, credits_consumed, credits_left. Wire video into KIE • Preview Video or KIE • Save Video; the node isn't a terminal node, so an unwired video output means it never executes.

Install and key

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

Or Manager → search "KIE.ai Nodes Next". Requirements are requests, Pillow, numpy, PyYAML; no weights, no VRAM, nothing to download. API key goes in once at Settings → KIE.ai Nodes Next → Connection - validated before saving, stored outside the workflow, masked to a fingerprint in the UI.

Where people get burned

Using an old task ID is the top one. Extensions are tied to a generation that still exists on KIE's side; IDs from a workflow you archived last month may simply not resolve, and the failure arrives as a provider error, not a friendly one.

Second: extending a clip you didn't generate through KIE. The field wants a KIE task ID, not a URL and not a local file. If you have a video you love from elsewhere, you can't extend it here - that's a different operation.

Third: cost discipline. Every extension is another full-price job, and stacking four of them to reach an arbitrary length costs more than shooting the sequence as two planned 8-second shots and cutting them together. Plan the beats first, then decide whether you need extension at all. The 4K-capable tiers aren't the place to discover that.