Video model ${model.modelId} does not implement doStart. Use

Error message

Video model ${model.modelId} does not implement doStart. Use generateVideo for models without an asynchronous start/status flow.

What it means

experimental_startVideo requires a model implementing doStart (the async start/status flow). If the resolved model has no doStart, it cannot produce an operation; the SDK throws and points you to generateVideo, which handles models without an async flow.

Source

Thrown at packages/ai/src/generate-video/start-video.ts:135

  seed?: number;
  frameImages?: Array<{
    image: DataContent;
    frameType: Experimental_VideoModelV4FrameType;
  }>;
  inputReferences?: Array<
    DataContent | { data: DataContent; mediaType?: string }
  >;
  generateAudio?: boolean;
  providerOptions?: ProviderOptions;
  maxRetries?: number;
  abortSignal?: AbortSignal;
  headers?: Record<string, string>;
  webhookUrl?: string;
}): Promise<StartVideoResult> {
  const model = resolveVideoModel(modelArg);

  if (model.doStart == null) {
    throw new Error(
      `Video model ${model.modelId} does not implement doStart. ` +
        'Use generateVideo for models without an asynchronous start/status flow.',
    );
  }

  if (!Number.isInteger(n) || n < 1) {
    throw new Error(
      `Invalid n: expected a positive integer, received ${JSON.stringify(n)}.`,
    );
  }

  // A start yields one operation covering all n videos: refuse to silently
  // exceed a known per-call limit instead of splitting into several starts.
  const knownMaxVideosPerCall =
    maxVideosPerCall ??
    (typeof model.maxVideosPerCall === 'function'
      ? await model.maxVideosPerCall({ modelId: model.modelId })
      : model.maxVideosPerCall);

View on GitHub (pinned to 69428b1f8b)

Solutions

Switch to a model that supports the async start/status flow.
Use generateVideo instead for synchronous models.
Implement doStart in your custom model wrapper if you own it.
Gate the call at runtime by checking typeof model.doStart === 'function'.

Example fix

// before
const op = await experimental_startVideo({ model: syncModel, prompt });
// after
const { video } = await generateVideo({ model: syncModel, prompt }); // sync model
// or: const op = await experimental_startVideo({ model: asyncModel, prompt });

Defensive patterns

Strategy: type-guard

Validate before calling

if (model.doStart == null) {
  // use generateVideo for this model instead
}

Type guard

function supportsDoStart(model) {
  return typeof model?.doStart === 'function';
}

Try / catch

try {
  await experimental_startVideo({ model, prompt, n });
} catch (e) {
  if (e.message.includes('does not implement doStart')) {
    const { video } = await generateVideo({ model, prompt, n });
  }
}

Prevention

Route models without async support to generateVideo.
Verify doStart exists on custom model implementations.
Maintain a capability map of your models.
Update provider packages to versions with async video support.

When it happens

Trigger: Calling experimental_startVideo (or startVideoStep) with a video model whose provider only supports synchronous generation (doStart undefined).

Common situations: Using startVideo against a synchronous video model; custom model implementation missing doStart; migrating code from generateVideo to startVideo without checking model support.

Related errors

AI-assisted analysis of vercel/ai@69428b1f8b (2026-08-30). Data as JSON: /api/errors/bb00ce37112d8a1e. Report an issue: GitHub.