Videos

Generate videos in bulk with batch.videos — Grok Imagine prompts in, signed video URLs out, with editing and extension through provider options.

Submit a video batch

batch.videos() mirrors batch(): pass a video model and a list of prompts, and it returns the same BatchJob handle immediately. Each request produces one video, correlated by customId.

import { batch } from "batchwork";
import { xai } from "@ai-sdk/xai";

const job = await batch.videos({
  model: xai.video("grok-imagine-video"),
  requests: [
    { customId: "a", prompt: "A red bicycle rolling downhill.", duration: 5 },
    {
      customId: "b",
      prompt: "A paper boat drifting down a stream.",
      duration: 5,
    },
  ],
});

const results = await job.wait().then(() => job.collect());
for (const r of results) {
  console.log(r.customId, r.videos?.[0]?.url);
}

Everything on the job handle works unchanged — wait(), poll(), results(), collect(), and cancel() — as does rehydration with getBatch / getBatchResults / cancelBatch. You can also pass the "xai/grok-imagine-video" string form.

Supported providers

xAI is currently the only provider whose batch API accepts video generation. OpenAI’s Videos API (Sora) is deprecated (shutting down September 24, 2026), so Batchwork deliberately doesn’t support it; Google’s Veo models aren’t batch-compatible. Other providers throw UnsupportedProviderError before any network request.

Request shape

Each request is a text prompt plus an optional customId and a few generation knobs:

limits and metadata work exactly as in batch(), and defaults merge into every request (request-level values win).

Editing and extending videos

xAI’s batch API also accepts video edits and extensions, and Batchwork routes each request line to the right endpoint automatically. The options mirror the AI SDK’s generateVideo provider options: pass a source videoUrl for an edit, or mode: "extend-video" to continue a clip. A single batch can mix generate, edit, and extend lines:

await batch.videos({
  model: xai.video("grok-imagine-video"),
  requests: [
    { customId: "gen", prompt: "A red bicycle rolling downhill." },
    {
      customId: "edit",
      prompt: "Make the bicycle blue.",
      providerOptions: { xai: { videoUrl: "https://example.com/source.mp4" } },
    },
    {
      customId: "extend",
      prompt: "The bicycle reaches the bottom of the hill.",
      providerOptions: {
        xai: {
          mode: "extend-video",
          videoUrl: "https://example.com/source.mp4",
        },
      },
    },
  ],
});

Reference-image guidance works the same way (providerOptions.xai.referenceImageUrls). Source URLs must be reachable by xAI while the batch processes.

Results

Videos reuse the normalized BatchResult. Generated videos land on result.videos, an array of { url?, durationSeconds? }:

import { writeFile } from "node:fs/promises";

for await (const result of job.results()) {
  const video = result.videos?.[0];
  if (result.status === "succeeded" && video?.url) {
    // Signed URL — expires ~1h after completion.
    const response = await fetch(video.url);
    await writeFile(
      `${result.customId}.mp4`,
      Buffer.from(await response.arrayBuffer())
    );
  } else if (result.status === "errored") {
    console.error(result.customId, result.error?.message);
  }
}

text, embedding, and images are undefined for video batches; the raw provider payload is on result.response.

How it’s built

Like batch(), batch.videos() derives each provider request body by running the AI SDK’s experimental_generateVideo() through a capturing fetch — the provider’s first network call is the JSON job-creation POST (polling only starts afterwards), so the captured body is exactly what a batch line needs. Edit and extend requests capture against their own endpoints (/v1/videos/edits, /v1/videos/extensions), which Batchwork writes per line.