grok Imagine Video 1.5

Grok Imagine Video 1.5 is a video generation model developed by SpaceXAI (formerly xAI) that produces cinematic video clips with native synchronized audio. Unlike models that apply audio in a separate post-processing step, this model generates sound effects, ambient sound, and lip-synced dialogue in a single inference pass alongside the visual content. It is primarily used for image-to-video (I2V) and text-to-video (T2V) workflows, allowing users to animate still frames or generate entire scenes from natural language descriptions.

The model is built on the Aurora engine, an autoregressive mixture-of-experts (MoE) architecture. By predicting tokens across interleaved modalities, the engine maintains high temporal consistency and physical realism, handling complex dynamics such as fluid motion, lighting reflections, and material translucency. The model was trained on the Colossus supercomputer cluster, leveraging high-density hardware configurations to optimize rendering speeds; a standard 6-second 720p clip typically processes in approximately 25 seconds.

Key Capabilities

• Reference-Guided Generation: Users can provide up to seven reference images to lock in specific characters, locations, or product identities, ensuring visual consistency across multiple shots.
• Native Audio Synchronization: Generates high-fidelity audio that matches the movement and pacing of the video, including clear speech for character dialogue.
• Video Extension: Allows for the expansion of existing clips by using the final frame of a generation as the starting point for the next, enabling the creation of longer sequences.
• Resolution and Aspect Ratios: Supports multiple output resolutions including 480p, 720p, and 1080p across seven common aspect ratios such as 16:9, 9:16, and 1:1.

Prompting and Scene Direction

Effective prompting for Grok Imagine Video 1.5 utilizes natural language to direct camera movement, atmosphere, and sound design. Users are encouraged to describe specific camera actions—such as "slow cinematic push-in" or "gentle parallax glide"—to control the perspective. Including descriptions of the desired soundscape within the prompt, such as "low rumbling bass" or "crisp footsteps on gravel," helps guide the native audio generation to match the visual intent. For character-focused clips, providing a high-quality portrait as an input reference significantly improves facial accuracy and reduces motion artifacts.