Grok Imagine 1.0: AI Video Generator Technical Reference
Grok Imagine 1.0 technical reference: specs, grok image to video capabilities, and what changed from preview to GA. See how 1.0 compares to 1.5.
Before reading, try current Grok Imagine video generation.
Developer: xAI · Architecture: Aurora (Autoregressive MoE) · GA Released: February 2, 2026 · API Announced: January 28, 2026 · API model ID: grok-imagine-video
Version note: This page covers Grok Imagine 1.0, the first general availability release of xAI's AI video generator. For the complete model family guide including Aurora architecture and image generation, see the Grok Imagine complete guide. For the current recommended version, see Grok Imagine 1.5.
What Made Grok Imagine 1.0 Significant
Grok Imagine 1.0, released on February 2, 2026, marked the first time xAI's video generation capability moved beyond the X platform ecosystem and became accessible to third-party developers through a public API. The January 28, 2026 API announcement described it as "state-of-the-art video generation across quality, cost, and latency — a unified bundle of powerful APIs designed for end-to-end creative workflows."
The 1.0 release represented two converging milestones: the first general availability of Aurora-based video generation for API consumers, and the graduation of a product that had been in staged rollout on X since August 2025. During that rollout period, the model reportedly generated a high volume of video content for X subscribers, reflecting the scale of usage built up during the platform-native preview phase before the API opened to the broader developer community.
The release also established Aurora as a commercially viable autoregressive video generation architecture at a time when the market was largely dominated by diffusion-based models. Before 1.0, every major video generation model available on API (Sora, Veo, Kling, Seedance) used diffusion. Grok Imagine 1.0 was the first autoregressive system to reach broad API availability.
From Preview to 1.0: What Changed
Grok Imagine's video generation had been in staged rollout since August 2025, initially limited to X Premium subscribers on iOS. The 1.0 release on February 2, 2026 was a genuine capability upgrade from the preview version, not simply a re-labeling:
Clip duration extended. Maximum single-pass generation increased from 6 seconds to 10 seconds, enabling more complete scene development within a single generation.
Resolution upgraded. Output resolution advanced from 480p to 720p (1280×720), bringing Grok Imagine to the resolution tier used by most professional content platforms.
Audio quality substantially improved. Native audio (dialogue, ambient sound, and sound effects) became significantly more natural in 1.0 than in the preview version, where audio quality was frequently cited in early user feedback as a limitation.
API access opened. For the first time, third-party developers and platforms could access Grok Imagine video generation without requiring an X subscription. The xAI API launched alongside the consumer 1.0 release.
Text-to-Video in Grok Imagine 1.0
Text-to-video (T2V) in Grok Imagine 1.0 generates clips directly from natural language prompts. The model produces 720p output at 24 FPS with native audio synchronized to the visual content. Clip duration ranges from 6 to 10 seconds per generation.
Three creative modes are available:
- Normal: Standard generation mode for most content types
- Fun: Stylized, heightened, or exaggerated visual treatment
- Spicy: Mature content mode; availability depends on platform content policy
The model accepts prompts up to 5,000 tokens, providing substantial space for detailed scene direction, camera movement specification, character description, and audio cues.
Effective prompt structure for T2V: Scene description → subject and action → camera behavior → atmosphere and lighting → audio direction.
Example: "Coastal cliff at dusk, handheld camera with slight drift. A woman in her forties looks out at the ocean, hair moving in wind. She turns slowly toward camera. Golden light from the west, lens flare on the turn. Wave sounds, distant seagulls, quiet wind."
T2V does not benefit from Aurora's hard first-frame anchoring; that advantage applies specifically to I2V, where a source image provides the literal starting frame. For T2V, the model generates all frames from the text description without a visual anchor, which produces weaker subject identity preservation compared to I2V.
Grok Image to Video in Grok Imagine 1.0
Image-to-video (I2V) is architecturally Grok Imagine's strongest video generation mode, and this applies fully to version 1.0. The source image is not treated as a style reference or conditioning signal. Aurora treats it as the literal first frame of the generation sequence, continuing the token prediction from that anchor point.
How I2V works in 1.0:
Provide a still image as input alongside a text prompt describing the desired motion, camera behavior, and audio. The model generates a video clip with the source image as frame 1. Subject identity, facial geometry, lighting, composition, and color carry directly into the generated frames because the generation is continuing from the source image's own token sequence, not approximating it from a description.
Key I2V capabilities in version 1.0:
- Clip duration: Up to 10 seconds at 720p at 24 FPS
- Aspect ratio: Output aspect ratio specifiable across 7 supported formats, independently of the input image's native ratio
- Native audio: Dialogue, ambient sound, SFX, and BGM generated in the same pass as the video
- Single-character lip sync supported
Best results with I2V: Portrait animation (consistent facial identity), product animation (color and branding accuracy from the first frame), and any use case where the source image defines the visual identity that must persist throughout the video.
Reference-to-Video in Grok Imagine 1.0
Reference-to-Video is a separate generation mode available in grok-imagine-video (1.0) that accepts up to 7 reference images via @image1…@image7 syntax. Unlike standard I2V, which treats the source image as the literal first frame, Reference-to-Video uses the provided images as visual conditioning references for character, style, or object consistency throughout the generated clip — without locking the video to a specific starting frame.
For use cases requiring multiple input images to establish multiple characters, environments, or props in a single generation, Reference-to-Video is the appropriate mode. Note: this mode is not available in Grok Imagine 1.5. For use cases where no source image exists at all, T2V is the appropriate mode.
Output Specifications
| Specification | Grok Imagine 1.0 (grok-imagine-video) |
|---|---|
| Architecture | Aurora (Autoregressive MoE) |
| Clip duration | 6–10 seconds |
| Resolution | 480p · 720p |
| Frame rate | 24 FPS |
| Aspect ratios | 16:9 · 9:16 · 4:3 · 3:4 · 2:3 · 3:2 · 1:1 |
| Audio | Native — dialogue · SFX · ambient · BGM |
| Lip sync | Yes (single character; best results) |
| Creative modes | Normal · Fun · Spicy |
| Max reference images (Reference-to-Video) | Up to 7 |
| Context window | 5,000 tokens |
| Generation speed | Slower than 1.5 Fast; expect a tens-of-seconds runtime for a standard 10-second 720p clip |
Source: xAI Grok Imagine API Announcement · xAI API Documentation · As of July 2026
How Grok Imagine 1.0 Compared at Release
At the time of its February 2026 release, Grok Imagine 1.0 entered a competitive landscape that included Sora (OpenAI), Veo (Google DeepMind), Kling (Kuaishou), and Seedance (ByteDance). Its competitive position was defined primarily by:
vs. Sora (OpenAI): Grok Imagine 1.0 offered faster generation times and a lower API cost than Sora's Quality tier. Sora maintained advantages in longer clip duration and broader resolution support. (OpenAI subsequently discontinued the standalone Sora 2 application in April 2026.)
vs. Veo 3.1 (Google DeepMind): Veo 3.1 led on resolution (4K), Scene Extension for longer-form content, and audio synchronization precision in multi-person dialogue. Grok Imagine 1.0 competed on generation speed and Aurora's hard first-frame I2V anchoring, which produced stronger subject identity preservation than Veo's conditioning approach.
vs. Kling 3.0 (Kuaishou): Kling 3.0 offered substantially longer single-pass clip generation and native 4K output, making it structurally stronger for long-form content. Grok Imagine 1.0's advantage was in generation speed for short-form content and I2V subject identity fidelity.
vs. Seedance 2.0 (ByteDance): Seedance 2.0 offered richer multimodal reference inputs and stronger physical realism in complex motion. Grok Imagine 1.0 competed on I2V subject identity preservation via hard first-frame anchoring.
For current competitive context reflecting the mid-2026 landscape, see Grok Imagine 1.5.
Known Limitations in Grok Imagine 1.0
10-second maximum clip duration. For content beyond 10 seconds, clips must be assembled from multiple generations or extended using the Extend from Frame feature (added March 2, 2026).
720p maximum video resolution. For workflows requiring 1080p or higher final output, Grok Imagine 1.0 is not the appropriate choice. Version 1.5 added 1080p support.
T2V subject identity coherence. Without a source image to anchor from, text-to-video generates subject appearance from the text description, producing weaker identity consistency across the clip. For any subject that must appear consistently, I2V is the structurally correct starting point.
Single-character lip sync. Multi-person simultaneous dialogue produces more variable lip-sync accuracy. Most reliable results are with one speaking character per clip.
Pricing. Version 1.0 is priced lower per second than 1.5 at both resolutions. For high-volume pipelines, accumulated cost per clip is a factor worth planning for. See xAI API documentation for current rates.
Developments After the 1.0 Launch
xAI continued updating the Grok Imagine model family after the 1.0 GA release:
Extend from Frame (March 2, 2026). xAI introduced the ability to chain video clips by extending from the final frame of an existing generation. This addressed the 10-second single-pass duration limit, enabling sequences up to approximately 30 seconds through chaining. The feature used the grok-imagine-video (1.0) endpoint and is the same mechanism refined and labeled as Video Extend in version 1.5.
Quality Mode API (May 6, 2026). xAI announced a Quality Mode update adding new output options and resolution improvements, marking a significant incremental update to the 1.0 product line ahead of the 1.5 release.
Upgrading to Grok Imagine 1.5
Grok Imagine 1.5, released in general availability on June 16, 2026, is the recommended upgrade path from 1.0. The migration is API-compatible. The model identifier changes from grok-imagine-video to grok-imagine-video-1.5, but prompt structure, reference image handling, and output parameters are backward compatible. No prompt tuning or workflow restructuring is required.
Key improvements in 1.5 over 1.0:
- Maximum clip duration: 10 seconds → 15 seconds
- 1080p resolution support added
- +52 Elo point improvement on the Image-to-Video Arena leaderboard (reaching #1)
- Fast tier: ~25 seconds for a 6-second clip
- Improved audio synchronization, motion quality, and temporal coherence
- Better Video Extend continuity across clip boundaries
Pricing change to factor in: Version 1.5 is priced higher per second than 1.0 at both 480p and 720p. The 1.5 480p tier can partially offset the difference in cost-sensitive workflows that do not require 720p output. See xAI API documentation for current rates.
For the complete 1.5 guide including Video Extend improvements, speed tier details, and full pricing, see Grok Imagine 1.5.
Frequently Asked Questions
When did Grok Imagine 1.0 launch? The general availability release was February 2, 2026. The xAI API was announced on January 28, 2026, giving developers access ahead of the consumer launch.
What was the difference between the preview and 1.0? The preview version (August–January 2025–26) was limited to X Premium subscribers on iOS, capped at 6-second clips at 480p, and had audio quality limitations. Version 1.0 extended duration to 10 seconds, upgraded resolution to 720p, substantially improved audio quality, and opened API access to third-party developers for the first time.
How fast does Grok Imagine 1.0 generate video?
It is generally slower than Grok Imagine 1.5 Fast and should be expected to take tens of seconds for a standard 10-second 720p clip on the grok-imagine-video endpoint. This reflects Aurora's autoregressive sequential generation process.
What is the grok-imagine-video API model ID?
grok-imagine-video is the API model identifier for Grok Imagine 1.0. grok-imagine-video-1.5 is the identifier for version 1.5. Both are accessible through the xAI API with a valid API key.
Is Grok Imagine 1.0 still available?
Yes. The grok-imagine-video endpoint remains available through the xAI API as of July 2026. Version 1.5 (grok-imagine-video-1.5) is available at a higher per-second rate. See the xAI API documentation for current pricing.
Should I still build on Grok Imagine 1.0?
For new workflows, Grok Imagine 1.5 is recommended. The 1.0 endpoint may still be appropriate if you have existing workflows on the grok-imagine-video endpoint not yet migrated, or if the lower per-second cost at 720p is the primary decision factor for your use case.
Grok Imagine is available on VioEvo, supporting image-to-video and text-to-video workflows.
