In January 2026, Grok Imagine generated 1.245 billion videos. Let that number settle for a moment. One billion, two hundred and forty-five million videos — in a single month, from a tool that launched its first rough iteration less than a year earlier. The scale of adoption tells you everything about where video generation is heading, and where xAI has positioned itself in that race.
Grok Imagine is xAI's image and video generation suite, powered under the hood by the Aurora autoregressive engine — a proprietary model trained on a cluster of 110,000 NVIDIA GB200 GPUs. It is accessible through the Grok app on iOS, Android, and web, and through the xAI API for developers. Here is a complete picture of everything it can do in 2026.
The timeline: how Grok Imagine got here
Grok Imagine launched in July 2025 with six-second text-to-video clips and audio — rough, but functional. By October 2025 (v0.9), generation time had been cut to under 15 seconds, and instant image generation was added. On January 28, 2026, the API launched with text-to-video, image-to-video, and video editing support. February 3, 2026 brought version 1.0: video length extended to 10 seconds at 720p with dramatically improved audio and prompt-following accuracy. March 2 added Extend from Frame. March 4 added folder organization. Late March brought version 1.3.54 with sharper detail, cinematic flair, and smoother motion. April 3 introduced speed and quality mode selection. Imagine 2.0 — with major face consistency and audio improvements — is confirmed as weeks away.
Text-to-video: what you can make today
The core workflow is straightforward: describe a scene in natural language, and Grok Imagine generates a video clip from scratch. Current specifications: up to 10 seconds per clip at 720p resolution with synchronized audio. Generation takes under 15 seconds for most prompts. The Aurora engine's headline capability is instruction-following accuracy — complex compositional prompts with specific lighting, camera movement, character positioning, and style requirements are handled with a level of precision that independent benchmarks confirm as competitive with the leading models.
The types of content that work best: cinematic scene descriptions with explicit lighting and camera direction, stylized sequences (anime, cyberpunk, fantasy), product-style shots with controlled environments, and social content with a distinctive visual identity. The types of content that still struggle: complex multi-object physical interactions, precise anatomy across motion, and long-form narrative continuity beyond a single clip.
Image-to-video: animate what you already have
Upload a still image — your own photo, an AI-generated image, a product shot — and Grok Imagine animates it. The model preserves the original visual identity while adding motion, atmosphere, and camera movement. This is one of the most practical creative workflows in Grok Imagine, because it gives you direct control over the visual starting point rather than relying entirely on the model's interpretation of a text description.
Extend from Frame: building longer sequences
The March 2, 2026 update introduced Extend from Frame — the feature that changed what Grok Imagine is capable of for serious creative workflows. Before this update, every generation began from zero. After, you can chain clips: the final frame of one generation becomes the starting point of the next, preserving motion, lighting, character positioning, and visual style across the transition.
Each extension generates a new clip up to 15 seconds at 720p. In theory, you can build multi-clip sequences of arbitrary length. In practice, community testing has confirmed that video quality degrades visibly after two or three chained extensions — a limitation xAI has acknowledged but not yet fixed with a timeline. For social content and short creative sequences, the limitation is not a blocker. For longer commercial production work, it is a real constraint to know before committing to a workflow.
Speed and quality modes — and what's coming next
As of April 3, 2026, Grok Imagine now offers a choice between Speed mode and Quality mode for image generation. Speed mode prioritizes fast output for rapid iteration through prompt variations. Quality mode dedicates more compute for higher-fidelity results. A Professional mode is confirmed for later in April, ahead of the full Imagine 2.0 rollout.
Imagine 2.0 is confirmed as a few weeks away as of early April 2026. The headline improvements: dramatically better face and detail consistency — the most visible weakness in AI-generated video across all providers — and improved speech-audio synchronization. These are the two improvements that would move Grok Imagine from a compelling creative tool to a serious production option for commercial workflows.
The API: pricing and developer access
The Grok Imagine API launched January 28, 2026 at api.x.ai. Pricing: $0.05 per second for 720p video with audio — approximately $0.50 for a 10-second clip, or $4.20 per minute of generated content. This is competitive relative to other video generation APIs at equivalent quality. The API supports text-to-video, image-to-video, and video editing workflows, and is available to all developers with a standard xAI API key. Templates including Chibi (anime-style) and other style presets are available for users who want consistent output without crafting custom prompts.
Access requirements
All video generation features — including Extend from Frame and the latest quality improvements — require an X Premium subscription. Basic X Premium starts at $8/month. SuperGrok and Premium+ subscribers receive more daily generations and higher-quality output. The Grok app update (version 1.3.54 or later) is required for the speed and quality mode selector and the most recent cinematic improvements — update via the App Store or Google Play if you have not already.