Nodes/comfyui-evolink/EvoLink Kling O3 Reference-to-Video (Official)
ComfyUI Node

EvoLink Kling O3 Reference-to-Video (Official)

Reference video in, new video out — Kling O3 steals style and motion

By deeplearning-goethe·Created 24 days ago·Updated a day ago· 0
EvoLink Kling O3 Reference-to-Video (Official)
  • reference_image1
  • reference_image2
  • video
  • task_id
  • status_info
  • result_urls
  • response_json
prompt
video_url
duration5
aspect_ratio16:9
quality720p
keep_original_soundtrue
api_key
timeout_seconds1200

EvoLink_KlingO3Ref is the reference-to-video node in the Kling O3 family. Where the image-to-video node animates from a still, this one takes an entire reference video as its source and extracts its style and motion to drive a brand-new clip, with optional reference images bolted on for subject control. Think of it as "make something new in the same groove as this clip." It's the node you use when you want a video's energy - camera behavior, pacing, motion character - without copying the video itself.

The reference video is a URL, and it's required in practice even though the field sits in the optional bucket: no reference video, no motion to transfer.

The inputs that matter

  • prompt (required) - shot and motion description, up to 2500 characters, Chinese-friendly. This is your creative direction on top of the reference.
  • video_url - the reference video. Wire it from an upstream EvoLink node's result_urls, or paste any public URL. This is the soul of the node.
  • keep_original_sound - defaults to on, carrying the source video's audio over.
  • reference_image1, reference_image2 - optional images to steer the subject (identity, look).
  • duration - 3–10 seconds, default 5.
  • quality - 720p (default) or 1080p.
  • aspect_ratio - 16:9 (default), 9:16, or 1:1.

Plus the pack-standard api_key and timeout_seconds (1200).

How it works

The reference images upload through EvoLink's file channel (auto-deleted after 72h, no third-party host), and the job posts to POST /v1/videos/generations as kling-o3-reference-to-video. The pack polls until the new clip completes and downloads it as a VIDEO object - connect video to a native Save Video node. Outputs: video, task_id, status_info, result_urls (24h expiry), response_json.

Installing

No GPU, no model files, only requests:

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

then fully restart ComfyUI, or search EvoLink in ComfyUI Manager.

Troubleshooting

401/402/403/429 plus the VPN note - whitelist api.evolink.ai and files-api.evolink.ai. The reference-node-specific trap is the 24-hour result link expiry: reference a stale result_urls from an old run and the job can fail at submission. If the output feels like it ignored the reference, check that video_url is actually populated - an empty field silently changes what the model gets. And keep the reference clip reasonably clean: muddled or heavily-cut source videos give muddled transfers.

CategoryEvoLink

Inputs (10)

NameTypeDefaultDescription
promptSTRING镜头/画面描述(必填),支持中文,最长 2500 字符。
video_urloptSTRING参考视频 URL(必填),可接上游节点的 result_urls 输出
durationoptINT53–10视频时长(秒),3~10 任意整数
aspect_ratiooptCOMBO16:9画幅比例
qualityoptCOMBO720p分辨率
keep_original_soundoptBOOLEANtrue保留原视频声音
reference_image1optIMAGE参考图 1(可选)
reference_image2optIMAGE参考图 2(可选)
api_keyoptSTRINGEvoLink API Key(sk- 开头)。首次填写后自动保存到本机配置,之后可留空。分享工作流前请清空此框。获取:evolink.ai/dashboard/keys
timeout_secondsoptINT120060–3600最长等待时间;生成失败或审核拦截的任务不扣费

Outputs (5)

NameTypeDescription
videoVIDEO生成结果(已下载落地,可直连保存节点)
task_idSTRINGEvoLink 任务 ID,可到 evolink.ai/zh/dashboard/logs 查询
status_infoSTRING人读状态摘要(模型/用时/消耗 credits)
result_urlsSTRING结果原始链接,每行一个(24 小时过期)
response_jsonSTRING平台 GET /v1/tasks/{id} 的完整 JSON 响应(status/usage/results 等),供下游节点解析