xAI's Grok Imagine Video 1.5 Lite arrives on Vercel AI Gateway

What launched

Vercel has added Grok Imagine Video 1.5 Lite, xAI's lightweight video generation model, to AI Gateway. According to the company's changelog, dated 8 October 2026, the model converts text prompts and images into video clips that come with native audio, so generated footage arrives with its own sound rather than requiring a separate audio step.

For developers already routing other model calls through the gateway, this is an additive change: video generation now sits behind the same interface as the rest of their AI traffic.

What the model produces

Per Vercel's changelog, the output envelope is:

• Clip length: 1 to 15 seconds
• Resolutions: 480p, 720p, and 1080p
• Seven aspect ratios spanning widescreen, square, and portrait formats

The aspect ratio coverage is the detail most likely to matter day to day. Because horizontal, square, and vertical outputs all come from one generation setup, a team building a clip-generation feature can serve a website hero video and a vertical social post from the same code path instead of maintaining separate workflows per format.

How developers use it

Vercel says generation starts by setting the model identifier to spacexai/grok-imagine-video-1.5-lite, either in the AI SDK or through the AI CLI. The changelog also points to a composition pattern: run an image model first, then hand the resulting still to Grok Imagine Video 1.5 Lite to animate it, treating image creation and animation as a single chained flow.

Anyone evaluating the output before writing code can try the model directly in the AI Gateway Playground.

The gateway layer underneath

The launch also illustrates how Vercel is positioning AI Gateway more broadly. The service offers a single API for invoking models, with usage and cost tracking built in, plus configurable routing, retries, and failover. Authentication is flexible: developers can use an AI Gateway API key or their own provider key, so teams holding existing xAI credentials do not have to re-provision access.

Why it matters

Video is the most operationally awkward generative modality: comparatively slow, comparatively expensive, and prone to failing partway through a request. Wrapping it in infrastructure that already handles retries, failover, and cost accounting removes a real chunk of the engineering between a weekend demo and something production-worthy. It also reflects a broader consolidation in model access, where text, image, and now audio-bearing video calls are becoming interchangeable line items behind one interface, making it cheaper to swap providers or run xAI alongside existing ones.

The Lite label is the other signal worth reading. Vercel describes the model as lightweight, and its 1080p ceiling and 15-second cap point to a tier tuned for speed and cost rather than maximal fidelity. That trade suits product surfaces like social clips, thumbnails, and animated stills, and the recommended pattern of chaining it after an image model suggests the expected use is animating existing imagery at least as much as generating elaborate scenes from a bare prompt.