Sign in
← AI PulseAug 20, 2026

Deep · news · Multi-source brief

xAI Introduces Reference-to-Video Generation with Grok Video Model

xAI has launched a new reference-to-video capability for its Grok video model, allowing users to generate videos guided by reference images, preset voices, or both.

By Illumora Editorial

Source · Aug 20, 2026, 3:13 PM · On Illumora · Aug 20, 2026, 4:04 PM

Media from the primary source — shown here so you can stay on Illumora.

Synthesized from multiple allowlisted primaries on the same event. Lanes →

Brief drafted by Illumora’s editorial model from the linked primary source. Ops desk reviews flagged pieces. How we write →

Read the source →xAI Docs — Reference-to-Video | SpaceXAI Docs
Save

xAI has introduced a new reference-to-video feature for its Grok video model, enabling the generation of videos using reference images, a preset voice, or a combination of both. This functionality is distinct from image-to-video generation, as it incorporates specific visual elements without locking the first frame of the output video.

Key Points

  • The Grok video model can generate videos using reference images, preset voices, or both.
  • Reference images can include specific people, objects, clothing, or other visual elements.
  • This method does not lock the first frame, differentiating it from image-to-video generation.
  • Users can provide up to 7 reference images per request.
  • Images can be supplied via public HTTPS URLs, base64-encoded data URIs, or file_id from the Files API.
  • The grok-imagine-video-1.5 model supports up to 3 optional preset voices selected by voice_id.
  • The maximum video duration on grok-imagine-video-1.5 is 15 seconds.
  • The maximum resolution for reference-to-video is 720p.

Context

According to xAI, the reference-to-video capability is designed for applications such as virtual try-on, product placement, character-consistent storytelling, and voice identity. The system requires at least one reference image or voice to initiate video generation. This feature cannot be combined with image-to-video or video editing within a single request, as only one mode can be active per request, determined by the parameters provided. For developers using the Vercel AI SDK, the providerOptions.xai.mode should be set to "reference-to-video", and images passed via providerOptions.xai.referenceImageUrls.

Files API integration, mentioned for providing file_id for reference images, allows users to manage and store files. The Files API is distinct from Collections, which provides persistent document storage with semantic search across multiple documents, according to xAI. The Grok CLI also supports a headless mode for scripting and automated tasks, as detailed by xAI, which could be relevant for programmatic video generation workflows.

Why It Matters

This new capability offers builders more granular control over video content generation, allowing for the consistent incorporation of visual and auditory elements across generated videos. It provides specific parameters and constraints that developers can use to guide their implementations, impacting how they design applications requiring custom video synthesis.

What To Do

  • Review the xAI documentation for the Grok video model to understand the full range of parameters for reference-to-video generation.
  • Test the reference-to-video feature in the Playground to observe its behavior with different combinations of reference images and preset voices.
  • Compare the capabilities and limitations of reference-to-video with image-to-video generation to select the appropriate mode for specific use cases.
  • Note the maximum limits for reference images (7) and video duration (15 seconds) when designing video generation workflows.

Keep Exploring

/atlas/**grok**-family /techniques/multimodal-grounding /studio?pack=foundation