EvoLink Seedance 2.5 Image-to-Video (Official)
First and last frame, 4 to 30 seconds of motion
- first_frame
- last_frame
- video
- task_id
- status_info
- result_urls
- response_json
Seedance 2.5 I2V is the image-to-video member of the current gen, and its two upgrades over 2.0 are the ones you'll actually feel: clips run up to 30 seconds, and a smart-duration mode lets the model time the shot. EvoLink Seedance 2.5 Image-to-Video (Official) is the closed API wrapper - required first_frame, optional last_frame, prompt in, VIDEO out. Nothing runs locally, because nothing about Seedance is downloadable.
What it is
- first_frame (required) - the starting image from any image output. Local files upload through EvoLink's channel and are auto-deleted after 72 hours. The clip's aspect ratio follows this frame.
- last_frame (optional) - the ending image. Connect it and you're in first/last-frame mode; mismatched ratios auto-align to the first frame.
- prompt (required) - the motion description, Chinese welcome, ~500 chars.
- duration - 4–30 seconds, default 5, billed by duration.
- auto_duration - the smart knob: it overrides
durationand lets the model pick a length in the 4–30 range, billed on what it actually outputs. - quality - 480p / 720p / 1080p. Note the top is 1080p; the 2.5 line dropped the 4K option that 2.0 I2V still has.
- generate_audio - on by default, no extra cost.
Outputs: video (into Save Video - nothing visible otherwise), task_id, status_info, result_urls (24-hour expiry), response_json. Default timeout 30 minutes; a long I2V render is a genuinely long wait.
Install and key
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
Dependency is just requests; no models, no GPU. Key from evolink.ai/dashboard/keys (sk-), pasted once - saved locally, blank after, never in a shared workflow JSON.
How to get the most out of it
The first/last-frame mode is the real editorial tool here, and 30 seconds makes it more useful: with a 5-second budget, first/last-frame was a tiny loop; at 30 seconds it's a scene with an arc. Use the loop trick (last frame = first frame) for long, seamless cycles, or pick a genuinely different end frame for a transition that has to physically resolve.
The one honest frustration is the resolution ceiling: if your hero frame deserves 4K motion, 2.5 I2V caps out at 1080p and you'll be eyeing the 2.0 I2V node instead. Otherwise the standing Seedance math applies - metered per render, prompt and frames leave the machine, drafts are cheaper on Fast/Mini or at 480p, and failed or moderation-blocked tasks don't charge. It's a great video node; just don't let it surprise you into a big bill.
Inputs (9)
| Name | Type | Default | Description |
|---|---|---|---|
| prompt | STRING | 镜头/画面描述,支持中文(建议 500 字内)。接 first_frame 出图生视频,可再接 last_frame 做首尾帧。 | |
| first_frame | IMAGE | 首帧图(必接) | |
| durationopt | INT | 54–30 | 视频时长(秒),4~30 任意整数,按时长计费 |
| auto_durationopt | BOOLEAN | false | 智能时长:开启后忽略 duration,由模型在 4~30 秒内自选,按实际输出时长计费 |
| qualityopt | COMBO | 720p | 分辨率;480p 为低价档 |
| generate_audioopt | BOOLEAN | true | 生成同步音频(不额外收费) |
| last_frameopt | IMAGE | 尾帧图(可选,接上即首尾帧模式;比例不同时自动对齐首帧) | |
| api_keyopt | STRING | EvoLink 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_secondsopt | INT | 180060–3600 | 最长等待时间;生成失败或审核拦截的任务不扣费 |
Outputs (5)
| Name | Type | Description |
|---|---|---|
| video | VIDEO | 生成结果(已下载落地,可直连保存节点) |
| task_id | STRING | EvoLink 任务 ID,可到 evolink.ai/zh/dashboard/logs 查询 |
| status_info | STRING | 人读状态摘要(模型/用时/消耗 credits) |
| result_urls | STRING | 结果原始链接,每行一个(24 小时过期) |
| response_json | STRING | 平台 GET /v1/tasks/{id} 的完整 JSON 响应(status/usage/results 等),供下游节点解析 |