This ComfyUI tutorial walks you through a compact text-to-video pipeline powered by the Wan3.0-Video model. The core of the workflow is the Wan3ReferenceToVideoApi node, which sends your prompt (and optional reference images) to the Wan3.0-Video service, then returns a single rendered clip. The SaveVideo node writes that clip to disk. You’ll control duration (up to 30 seconds), resolution/aspect, frame rate, seed, and other generation settings directly on the API node, making it easy to iterate from a single graph.
Technically, Wan3ReferenceToVideoApi handles the remote job lifecycle: it packages your text prompt and any reference inputs, submits the request, polls until completion, and outputs a finished video. The “Reference” capability helps maintain strong character or style consistency when you supply a face or character image. If your API plan/model supports it, the node can also return synchronized sound; otherwise you’ll receive a silent video you can score later. SaveVideo finalizes the output with your chosen filename pattern and fps, giving you a simple, reliable path from text to production-ready footage.
API
從程式碼使用此工作流
每個 Comfy 工作流都是一個 JSON 圖。下方的 payload 就是此工作流，與 ComfyUI 執行時完全相同 — 你可以從此 URL 取得、放入版本控制，或在 ComfyUI 載入並逐節點執行。
使用 Comfy SDK 以 TypeScript 或 Python 執行。相同程式碼可用於 Comfy Cloud 或你自架的 ComfyUI — 只需更改 base URL。
// Install (beta)
npm i @comfyorg/sdk
// Run this workflow (TypeScript)
import { Comfy } from "@comfyorg/sdk";
const client = new Comfy({ apiKey: "comfyui-..." });
const wf = await client.workflows.fromFile("workflow_api.json");
const job = await client.run(wf);
await job.getOutputs("<output-node-id>")[0].toFile("output.png");SDK 需使用 API 格式的工作流：請在 ComfyUI 開啟此工作流，並使用 檔案 → 匯出工作流（API）。
常見問題
