vidu

Vidu is the video engine from Shengshu AI, and this node is its door into ComfyUI - run entirely on Civitai's cloud, billed in Buzz, no local GPU involved. It's one of ~160 generated nodes in the official civitai-comfy-nodes pack, and it's the one with the party trick: you can give it both a start and an end frame, and Vidu animates between them.

That start-and-end capability is the reason to pick this node over its neighbors. Most video models take one image and run forward. Vidu takes "start here, end here" and does a journey. If you have a specific target composition - a character's final pose, a closing camera angle - this is the node that honors it.

The inputs that matter

Only prompt is required, but the optional inputs are where Vidu distinguishes itself:

• source_image - the start frame. Wire a LoadImage in. (The tooltip describes the underlying API field as "A URL, A DataURL or a Base64 string"; the node converts whatever IMAGE you give it into that for you.)
• end_source_image - the finish frame. This is the one to know about. Give Vidu a start and an end and you get a transition, not a random walk.
• style - general or anime. A genuinely useful switch if your content lives on the anime side; the general model doesn't do that aesthetic well.
• model - default, q1, or q3. Model family selector for the engine generation.
• movement_amplitude - auto, small, medium, large. How much motion you want: small for subtle, large for energetic scenes.
• duration - 4 or 8 seconds.
• aspect_ratio - 16:9, 9:16, or 1:1.
• enable_background_music - off by default; toggles generated music on the clip.
• enable_prompt_enhancer - on by default; server-side prompt rewrite. Turn off for exact prompts.
• seed, images, api_config - the usuals.

Outputs

Standard pack trio: video (VIDEO) into a Save Video / preview node, workflow_id (STRING) for the Civitai sidebar history, raw_json (STRING) to inspect the response and the per-run Buzz cost.

Install and auth

Shared one-time setup with the rest of the pack. ComfyUI Manager → Custom Nodes Manager → search Civitai Comfy Nodes → install, restart:

cd ComfyUI/custom_nodes
git clone https://github.com/civitai/civitai-comfy-nodes.git
pip install -r civitai-comfy-nodes/requirements.txt   # just `requests`

Authenticate via a Civitai Auth node, the CIVITAI_API_TOKEN env var (best for headless), or the sidebar sign-in.

Troubleshooting

• End frame ignored? If you wired end_source_image and the motion doesn't land on it, check enable_prompt_enhancer - a rewritten prompt can fight your explicit frames. Also make sure the end image's aspect ratio matches the start; mismatched shapes get cropped.
• Slow queues and timeouts. The pack went live June 2026 during Civitai's GPU crunch, and long waits were the top community complaint. Default timeout is 30 minutes; raise it on the Civitai Auth node or via CIVITAI_COMFY_TIMEOUT.
• Anime look with general style? Switch style to anime. This is the classic "why does this look wrong" on Vidu.
• It's metered. Background music and 8-second clips cost more. raw_json is where you see the meter; treat Vidu as a mid-tier cost engine - cheaper than Sora/Veo, pricier than Mochi.
• Early preview. The pack's README warns nodes "may change without notice." Check fields after updates.

Reach for this node when you have a beginning and an ending and want Vidu to fill in the middle. If you only have a start frame, the plain images input still works - but you're leaving its best feature unused.