Seedance 2.5 · Video Extend

You made four seconds you like. Now you need eight, or twelve, or a full thirty. Video Extend is the node that takes the clip you already have and generates the next segment from it, then hands back one continuous file with the new part already joined on. No stitching, no manual frame matching, no re-rolling and hoping the model finds the same shot again.

It's also the second-cheapest way to buy Seedance time in this pack - 11.07 credits per second (~$0.11/s) of new footage, with your source clip appended for free.

What the model is actually doing

Your clip is uploaded and the model reads its tail - for 2.5, up to the last 30 seconds of your source becomes reference footage for the new segment, and those reference seconds are billed on top of the output duration. Anything earlier in the clip is only joined back on; you're not paying for it as reference. That's a meaningful difference from the 2.0 version of this node, where the source was free and the last frame did all the work.

Practical consequence: extending a 40-second clip costs more than extending a 10-second one. Extending a 3-second clip costs less and gives the model less to go on.

Inputs worth setting

• video_url - MP4, MOV or WebM, 2 to 60 seconds. A VIDEO socket, so wire in a loader or the output of a save/load cycle.
• prompt - required here, unlike on the 2.0 extend node. Describe what happens next: the action, the camera move, the sound. Leaving it thin is the single biggest cause of a visible drift where the new segment stops looking like the old one.
• duration_seconds - 4 to 30, integer. Length of the new segment, and the number that gets multiplied by both the output rate and your reference seconds.
• resolution - 480p, 720p, 1080p. It only sets the quality of the new segment, so match your source or you'll get a soft-and-crisp seam in the same file.
• generate_audio - sound for the new segment only; your source keeps its own audio either way. On a dialogue shot this is where continuity gets shaky - the model isn't listening to the original track.
• enable_web_search - off. Leave it.
• watermark, seed - off, fixed.

Outputs

video (the complete extended clip - go straight to Save Video) and task id. There's no last frame output on this node; if you want to chain several extensions, save and reload.

Install

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

or install PopNow from ComfyUI Manager and restart. Add the key in Settings > PopNow (or POPNOW_API_KEY), restart a second time so your account's catalog builds the node list. No extra packages - this pack ships with an empty dependency list and uses only what ComfyUI already has.

Where it goes wrong

• PopNow 40004 on a source that looks fine - check container and length: MP4/MOV/WebM, 2–60 seconds.
• The seam. Two causes: resolution mismatch between source and resolution, or a prompt that asks for something the first segment was clearly not heading toward. Match the resolution, and write the continuation as if you were directing the same shot, not starting a new one.
• Drift on the third extension. Expected. Every hop is a fresh generation seeded by borrowed footage, so the look degrades a bit each time. Two or three hops is realistic; ten is a slideshow.
• PopNow 40201 - out of credits; remember reference seconds land on top of output seconds.
• Cancelling stops your wait, not the job. It runs to completion on PopNow, bills if it succeeds, and the file is in your API console.

Set Settings > PopNow > Max credits per run before you experiment. On a node whose price depends on two durations at once, a ceiling is the only reliable way to find out what a long extension costs without paying for the answer.