Vercel AI SDK Adds Support for xAI's Grok Imagine Video 1.5, Including 1080p Output
The @ai-sdk/xai package version 3.0.121 adds support for xAI's grok-imagine-video-1.5 model, including native 1080p resolution for text-to-video and image-to-video generation. The update also introduces referenceVoiceIds for reference audio and fixes a routing bug in reference-to-video mode.
What happened
Vercel shipped version 3.0.121 of @ai-sdk/xai, the xAI provider package for its AI SDK, adding support for xAI's grok-imagine-video-1.5 model. The patch release, tagged on GitHub, introduces native 1080p output for both text-to-video and image-to-video generation modes.
Technical details
According to the changelog, requests using the standard resolution: '1920x1080' parameter now map directly to the model's native 1080p output mode for text-to-video and image-to-video tasks. Reference-to-video generation — where a user supplies a reference image or clip — remains capped at 720p. If a developer requests 1080p in reference-to-video mode, the SDK downgrades the request and issues a warning rather than failing silently.
The update also fixes a routing bug affecting the inputReferences parameter. Previously, any non-empty inputReferences array triggered reference-to-video mode, even when the array contained only non-image references. This caused the SDK to send an empty reference_images: [] payload to xAI's API with no usable reference data, effectively breaking the request. The fix ensures reference-to-video mode only activates when a genuine image reference is present.
Separately, the release adds a new referenceVoiceIds parameter for reference-to-video generation. Developers can now pass up to three xAI preset voice IDs — for example ['eve'] — and reference them in prompts using placeholder tags <AUDIO_0> through <AUDIO_2>. The SDK translates these into a reference_audios array of { voice_id } objects sent to xAI's POST /v1/videos/generations endpoint, allowing generated videos to include synchronized preset voice audio.
Why it matters for developers
This is a patch release to Vercel's provider integration, not a new model announcement from xAI itself. The grok-imagine-video-1.5 model ID already existed on xAI's side; this update simply brings the AI SDK's parameter handling in line with the model's actual capabilities and fixes a real bug that silently broke reference-based video generation for some request shapes.
Pricing for grok-imagine-video-1.5 was not disclosed in this changelog. Developers using xAI's video generation API through Vercel's AI SDK should consult xAI's own API documentation for current per-generation or per-second pricing.
What this means
For teams building on xAI's video generation stack via Vercel's AI SDK, this update removes a rough edge: reference-to-video requests that previously failed due to the inputReferences misclassification bug should now route correctly. The 1080p support and voice-referencing feature also expand what's possible without dropping to xAI's raw API — useful for teams that want to stay inside the SDK's abstraction layer. This is routine maintenance work that keeps a third-party SDK synchronized with a provider's evolving API surface, rather than a signal of new model capability from xAI.
Related Articles
ElevenLabs Launches Music v2.5, Adds API Access and Free Tier for AI-Generated Songs
ElevenLabs has released Music v2.5, an updated version of its ElevenMusic generator, now available through both the app and API. The company says blind testing with nearly 48,000 comparison pairs showed listeners preferred v2.5 over the prior version, particularly for R&B, Hip-Hop, and orchestral genres.
Perplexity Says It Runs End-to-End Engineering Systems on OpenAI's GPT-6 Astra
Perplexity says it has shifted core engineering workflows, including code changes and production monitoring, onto OpenAI's GPT-6 Astra model. The claim comes from an OpenAI-published case study with no independent benchmark data released.
Perplexity Deploys OpenAI's Astra Model for Autonomous Code and Systems Management
Perplexity is using an OpenAI model referred to as Astra to handle software changes, communications, and production monitoring with less frequent human check-ins. OpenAI published the case study; specific model specs and benchmarks have not been disclosed.
Augment Code Claims 4.5x Developer Output Increase From Internal 'Software Factory' of AI Agents
Augment Code says its internal 'software factory'—a network of specialized agents built on its Cosmos platform—drove a 4.5x increase in size-adjusted developer output and cut median PR merge time from 11.2 to 3.1 hours over nine months. The company frames this as evidence that once AI writes nearly all new code, the bottleneck shifts to review, verification, and incident response.
Comments
Loading...