The Vercel AI SDK has been updated to include support for Grok Imagine Video 1.5, integrating the new model ID within its @ai-sdk/xai package. This update introduces enhanced video generation capabilities, particularly for text-to-video and image-to-video functionalities.
Key Points
- The @ai-sdk/xai package now supports the grok-imagine-video-1.5 model ID.
- Text-to-video and image-to-video generations can now utilize native 1080p resolution, with '1920x1080' mapping to this standard.
- Reference-to-video remains capped at 720p; 1080p requests in this mode are downgraded with a warning.
- A fix addresses reference routing, preventing empty
reference_images: []when non-image references are provided. - The SDK now supports
referenceVoiceIdsfor reference audio in reference-to-video. - Up to 3 xAI preset voice IDs can be passed, such as
['eve']. - These voice IDs are referenced in the prompt using
<AUDIO_0>–<AUDIO_2>and sent asreference_audios: [{ voice_id }]to the/v1/videos/generationsendpoint.
Context
According to the Vercel AI SDK changelog, these updates enhance the functionality of the @ai-sdk/xai provider. The changes include specific model support and resolution handling for video generation, alongside improvements in how reference inputs, including audio, are processed.
Why It Matters
This update provides developers using the Vercel AI SDK with direct access to Grok Imagine Video 1.5 capabilities, allowing for higher-resolution video outputs and more sophisticated control over reference audio in video generation tasks. This expands the creative possibilities for applications leveraging xAI's video models.
What To Do
- Review the @ai-sdk/xai documentation for updated model IDs and resolution specifications.
- Test text-to-video and image-to-video generations with 1080p resolution.
- Experiment with
referenceVoiceIdsto incorporate preset audio into reference-to-video projects. - Note the 720p cap for reference-to-video and adjust requests accordingly.
