EvoLink Seedance 2.5 Video Extend (Official)

Every video has a moment where it ends too early. EvoLink Seedance 2.5 Video Extend (Official) is ByteDance's answer to that: take an existing clip and keep it going - forward into what happens next, or backward into what led up to it - by 4 to 30 seconds of new footage that has to match. It's the sibling of the pack's video-edit node, and together they're the two halves of "don't regenerate, just fix it."

What it is

• prompt (required) - the extension instruction. The tooltip's own example uses a @video1 reference syntax: "Extend @video1 backward, 人物走出画面." That's the pattern - name the clip, say which way to extend it, describe the new action.
• video_urls - the clip(s) to extend, one URL per line, up to 10 (each 2–30 seconds). Marked required. As with the edit node, you can paste an upstream node's result_urls here directly - the generate-then-extend chain is the whole point.
• duration - the length of the extension, 4–30 seconds, default 5, billed by duration.
• quality - 480p / 720p / 1080p.
• generate_audio - on by default, free.
• audio_urls - optional reference audio, up to 10 entries.
• image1–image2 - optional image references (up to 30 with the URL box).

Outputs: video, task_id, status_info, result_urls (24-hour expiry), response_json. Save Video node on video or the extension evaporates into your credits history.

Install and key

The shared EvoLink install. ComfyUI Manager → search EvoLink → install → full restart (close the console window, not just refresh the tab). Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/deeplearning-goethe/comfyui-evolink

Only requests as a dependency; no models, no GPU. Key from evolink.ai/dashboard/keys (sk-), pasted once - saved locally, blank after, never in a shared workflow JSON.

The workflow move that makes it worth it

Extend is the "director's cut" node: generate a 5-second base, then chain result_urls into this node and extend it twice to build a 15-second sequence that keeps continuity, because each extension has to match the clip it's bolted onto. Backward extension is the sleeper feature - it lets you add context ("what led to this moment") without regenerating the part you already like.

The limits are worth naming before you commit credits: each extension is a separate metered render, so a long scene assembled from extensions costs like a long scene - which is fair, it is one - and output resolution tracks your input, so extend the good render, not the 480p test. Standing cloud deal applies: per-call billing, your clips and prompt leave the machine, failed or moderation-blocked tasks don't charge.