Zhenzhen Motion Video 动效视频 (Standard / Pro)
Zhenzhen Motion Video makes title cards, not films — and that's the point
- api_config
- video
- video_url
- task_id
- response
- cover_url
- preview_url
- html_preview_url
- subtitle_url
- storyboard
- fallback_scenes
It doesn't make footage
This node renders motion graphics: Canvas 2D animation, the stuff you'd otherwise build in After Effects or CapCut. Kinetic type, a 15-second hype intro, a data story, a logo sting wearing a template skin. The README says it outright - Canvas 2D work, "not a real-world video model." Want a clip of somebody walking down a street? Wrong node, same pack.
And none of it runs on your machine. This pack is a client for api.seedance.nz, the author's own paid gateway ("贞贞的平价AI小铺"). The model dropdown isn't a checkpoint, it's a remote endpoint, and every run is a metered call - the API-wrapper category, where you reach out because there's nothing to download.
What to set first
model gives you zhenzhen/motion-video or zhenzhen/motion-video-pro. Same node, better backend; try Standard short, and if the typography holds up you may never touch Pro.
template_id is the design, and the field I'd linger on. Past api_default there are ten looks: hype-kinetic, cyber-glitch, blueprint-tech, synthwave, swiss-minimal, code-cinema, data-story, chalk-hand, voxel-world, cinematic-trailer. Pick the aesthetic first, then write copy that fits it - fighting a template costs you calls.
The rest is a broadcast tool's shape: duration of 15/30/45/60 seconds, ratio at 16:9, 9:16 or 1:1, resolution at 720p or 1080p. prompt is the creative direction, validated as required before anything is submitted; the copy has its own fields - title, subtitle, keywords, signature - alongside language and music.
How it actually works
The node builds a payload with model, duration, ratio, resolution and prompt up top, then collects nearly everything else into a nested motion object: template_id, title, subtitle, signature, language, music, and keywords/assets as real arrays. Anything empty, or left as api_default, is dropped rather than sent. keywords and assets take one item per line or a JSON string array, and bad input is rejected locally before a call is spent.
Then it submits through the pack's legacy video route and polls. PREVIEW_READY in the console looks like success and isn't - the node keeps waiting through STORYBOARDING, MEDIA_GENERATING and RENDERING until the render is actually finished, which is correct and is where a lot of tools hand you a half-baked file.
The input that bites: assets
Every other node in this pack quietly uploads your local images. Motion Video is the deliberate exception. assets takes up to four persistent public HTTP(S) image URLs - one per line or a JSON array - and rejects anything with credentials embedded or no plain http(s) host. You need a real image host: a CDN, an S3 bucket, your own domain. A Load Image from the canvas is not accepted.
Second thing, from the author's own testing: all four URLs demonstrably reached the API and the reference images still rendered as geometric placeholders. The plumbing passes; the rendering doesn't. Treat assets as experimental until you've eyeballed an output.
Revising a finished cut
Set operation to revise and both revise_from (a completed Motion Video task ID - the previous node's task_id output wires straight in) and revision_instruction become required. Be specific about which scene: name-the-scene edits worked in testing, asking for a change to the default outro didn't.
Outputs
video is the downloaded file and the only one you'll normally wire - into Save Video or a preview. video_url, task_id and response are the pack's standard strings. cover_url, preview_url, html_preview_url, subtitle_url, storyboard and fallback_scenes fill out the rest; the last two arrive as JSON text, and anything the API omitted comes back as an empty string rather than breaking the node.
fallback_scenes is worth a glance. If upstream couldn't honour your storyboard it degrades shots and ships the video anyway - console warning, nothing failing in the graph. That field is the only thing telling you why shot three looks wrong.
Install and key
Manager: search ComfyUI Seedance, or comfy node install seedance. Manually:
cd ComfyUI/custom_nodes
git clone https://github.com/T8mars/ComfyUI_Seedance.git
cd ..
python -m pip install -r custom_nodes/ComfyUI_Seedance/requirements.txt
Requirements are one line, requests>=2.28. No torch, no CUDA, no model downloads. Restart and the node appears under Seedance. For auth, wire a Seedance API Config node (https://api.seedance.nz plus api_key) into api_config, or set SEEDANCE_API_KEY, or drop a config/.env in the pack folder.
Where people get burned
The seed is a cache key, not a sampler. The pack injects a standard seed control into every generation node, and Motion Video never sends it to the API. randomize re-executes and re-bills; fixed with nothing else changed can hand the same render back free - and explains why the node "ignored your edit" when you changed a field that isn't part of the cache key.
language defaults to zh-CN, as free text, not a dropdown. If your copy is English, say so.
Layout rejections happen. The author notes upstream layout validation can refuse a generated scene; the node then errors loudly instead of dressing a preview up as a finished video. Shorten the copy or switch template.
SSL failures on Windows portable installs are the pack's most common connectivity bug and they aren't a bad key - refresh the network deps in the Python that actually runs ComfyUI, and prefer SEEDANCE_CA_BUNDLE over SEEDANCE_SSL_VERIFY=0:
python -m pip install -U requests certifi
Result URLs expire. The *_url outputs are signed links on a timer; only the local video file persists. And keep a copy of the source handy - this pack is Chinese-language-first with almost no English footprint, so when a README line doesn't match behaviour, grep the class.
Inputs (19)
| Name | Type | Default | Description |
|---|---|---|---|
| model | COMBO | zhenzhen/motion-video | 2 options: zhenzhen/motion-video, zhenzhen/motion-video-pro |
| operation | COMBO | generate | 2 options: generate, revise |
| prompt | STRING | — | |
| duration | COMBO | 15 | 4 options: 15, 30, 45, 60 |
| ratio | COMBO | 16:9 | 3 options: 16:9, 9:16, 1:1 |
| resolution | COMBO | 720p | 2 options: 720p, 1080p |
| template_id | COMBO | api_default | 11 options: api_default, hype-kinetic, cyber-glitch, blueprint-tech, synthwave, swiss-minimal, +5 |
| title | STRING | — | |
| subtitle | STRING | — | |
| keywords | STRING | One keyword per line, or a JSON string array | 每行一个关键词,或 JSON 字符串数组 | |
| signature | STRING | — | |
| language | STRING | zh-CN | — |
| music | COMBO | auto | 2 options: api_default, auto |
| assets | STRING | Up to 4 persistent public HTTP(S) image URLs, one per line or JSON array; no local files | 最多4个持久图片直链,不接受本地文件 | |
| revise_from | STRING | Your completed Motion Video task ID | 自己已完成的动效视频任务ID | |
| revision_instruction | STRING | — | |
| api_configopt | SEEDANCE_CONFIG | — | |
| skip_erroropt | BOOLEAN | false | — |
| seedopt | INT | 00–18446744073709550000 | ComfyUI cache seed. Fixed reuses the cached result while all other inputs stay unchanged; randomize/increment/decrement starts a new execution. This value is not sent to models without documented seed support. | ComfyUI 缓存种子;Fixed 在其他输入不变时复用缓存,随机、递增或递减会触发新任务。未声明支持 seed 的模型不会收到此参数。 |
Outputs (10)
| Name | Type | Description |
|---|---|---|
| video | VIDEO | — |
| video_url | STRING | — |
| task_id | STRING | — |
| response | STRING | — |
| cover_url | STRING | — |
| preview_url | STRING | — |
| html_preview_url | STRING | — |
| subtitle_url | STRING | — |
| storyboard | STRING | — |
| fallback_scenes | STRING | — |