Nodes/ComfyUl-UCloud/Vidu StartEnd2Video
ComfyUI Node

Vidu StartEnd2Video

Vidu StartEnd2Video between two frames

By ucloud·Created about a year ago·Updated 3 months ago· 3
Vidu StartEnd2Video
  • client
  • first_frame_image
  • last_frame_image
  • url
  • task_id
modelviduq3-pro
duration5
resolution720p
movement_amplitudeauto
first_frame_url
last_frame_url
prompt
seed0
bgmfalse

Sometimes you don't want to describe a scene at all - you want the video to start on one image and end on another, with the model inventing the motion in between. That's Vidu StartEnd2Video, and it's the most direct tool in the Vidu pack for storyboard-style work: shot A transitions into shot B, and you control both ends of the transition.

This is the node for establishing shots, scene changes, object transformations ("this car turns into that car"), or any shot where the beginning and end are fixed and the interesting part is the middle. It pairs naturally with the pack's image nodes - generate your two keyframes, drop them in, and let Vidu fill the gap.

The inputs that matter

  • client - standard wiring from Modelverse Client or Secret Client.
  • model - five options: viduq3-pro (default) and viduq3-turbo (1–16s), plus viduq2-pro-fast, viduq2-pro, and viduq2-turbo (1–8s).
  • duration - 1–16s on q3 models, 1–8s on q2. The node checks this and refuses durations a model can't handle.
  • resolution - 540p, 720p (default), 1080p.
  • movement_amplitude - auto, small, medium, large - the motion dial for how much the camera or subject moves between the frames.
  • first_frame_image / first_frame_url - required, one or the other, never both.
  • last_frame_image / last_frame_url - also required, same rule. No first-frame-only here; this mode needs both bookends.
  • prompt - optional, max 2000 characters, for steering the transition.
  • seed and bgm (background music).

Outputs are url and task_id, with url going to Modelverse Preview Video.

How it works

The node base64-encodes both frames, submits the task with vidu_type: start-end2video, and polls every 5 seconds until done. One grounded gotcha lives in the source comments: the two frames should be similar in resolution (the code notes a ratio around 0.8–1.25). If your first frame is 1024×1024 and your last is a 512×512 crop, the model has to reconcile conflicting geometry and the transition will look rough. Resize both ends to a matching size before wiring them in.

Install

Pack install: ComfyUI Manager → search "ComfyUl-UCloud" → Install → restart, or git clone https://github.com/ucloud/comfyui-plugin into custom_nodes. Light dependencies, no downloads, API key from the Modelverse console.

The realistic note: transitions between two different subjects can drift - Vidu will happily morph things in ways that surprise you. If you want a clean, physical scene change, keep the two frames compositionally similar (same camera angle, similar framing) and let the prompt describe the transformation. Same-subject frames give the cleanest results; that's also why it's a good character-consistency trick when you only have two keyframes to work with.

CategoryUCLOUD_MODELVERSE/Vidu

Inputs (12)

NameTypeDefaultDescription
clientMODELVERSE_API_CLIENT
modelCOMBOviduq3-providuq3-pro/turbo: 支持1-16秒; viduq2-pro-fast/pro/turbo: 支持1-8秒
durationINT51–16视频时长(秒),viduq3系列支持1-16秒,viduq2系列支持1-8秒
resolutionCOMBO720p分辨率
movement_amplitudeCOMBOauto运动幅度
first_frame_imageoptIMAGE首帧图片
first_frame_urloptSTRING首帧图片URL
last_frame_imageoptIMAGE尾帧图片
last_frame_urloptSTRING尾帧图片URL
promptoptSTRING文本提示词,最长2000字符
seedoptINT00–2147483647随机种子
bgmoptBOOLEANfalse是否添加背景音乐

Outputs (2)

NameTypeDescription
urlSTRING
task_idSTRING