Grok Imagine 1.0: Complete Model Guide (Image & Video)
A complete guide to xAI's Grok Imagine 1.0 model, covering image and video generation capabilities, Aurora architecture, and performance analysis.
Grok Imagine 1.0: Complete Model Guide (Image & Video)
Developer: xAI · Architecture: Aurora (Autoregressive Mixture-of-Experts) · Video 1.0 Released: February 2, 2026 · Official xAI product page: Grok
Version note: This page covers Grok Imagine 1.0, the version currently available on our platform.
What Is Grok Imagine?
Grok Imagine is xAI's unified creative generation system — a single model family that handles image generation, image editing, text-to-video, and image-to-video in one integrated product. It is built on Aurora, xAI's proprietary autoregressive generation architecture, and is distinct from the Grok conversational AI: the two products share a brand but serve separate purposes.
The key distinction that makes Grok Imagine architecturally different from most competing models is Aurora's autoregressive foundation. Where most image and video generation models today use diffusion architectures, starting from noise and progressively denoising toward a final output, Aurora generates tokens sequentially, building the image or video frame through a series of predictions rather than a denoising process.
This architectural difference has a practical consequence: Aurora develops a deep understanding of both visual structure and the semantic meaning of content within that structure. In practice, that shows up as unusually strong text rendering inside generated images, prompt adherence that reflects genuine comprehension of compositional intent, and image-to-video conversion that treats the source image as a hard anchor for the first frame, producing subject identity preservation that diffusion-based models can struggle to match.
xAI built Aurora on the Colossus supercluster, a training infrastructure of 110,000 NVIDIA GB200 GPUs described at the time as the world's largest GPU farm. The scale of the training compute reflects a deliberate strategy: train a foundation model with enough capacity to handle the full spectrum of visual generation tasks within a single unified architecture, rather than building separate specialized models for images and video.
Part 1: Image Generation
What Aurora Produces
Grok Imagine's image generation converts text prompts into finished images at up to 2K resolution (2048×2048). The model handles a wide range of visual styles from a single endpoint, including photorealistic, anime, illustration, oil painting, 3D-rendered, and abstract, without requiring separate model selection for each. Style is specified in the prompt, not in a separate parameter.
Text rendering is a standout capability. The autoregressive architecture's token-sequential generation gives Aurora a structural advantage over diffusion models on this specific task. Logos, brand names, signage, titles, and multilingual copy, including Japanese, Korean, Chinese, and Arabic scripts, appear legibly inside generated images at a consistency level that diffusion-based generators still struggle to match reliably.
For creators building brand assets, marketing materials, posters, or any content where text accuracy inside the image matters, this is the capability that makes Grok Imagine the right tool rather than an acceptable one.
Image Generation Capabilities
Text-to-Image Generate images from text prompts with full control over aspect ratio, resolution, style, and output count. The model accepts prompts up to 10,000 characters, allowing for detailed compositional direction without prompt truncation.
Image Editing Upload an existing image and describe changes in natural language. Aurora edits with up to 3 reference images per request, enabling: background replacement, style transfer, object addition or removal, color and lighting adjustments, and character or costume changes. The editing workflow uses the same prompt syntax as generation, so there is no separate editing interface or parameter set required.
Batch Generation Both text-to-image and image editing accept a batch parameter (1–4 images per request for standard; up to 10 per request on Image Quality tier), enabling parallel candidate generation for creative exploration and A/B selection in production pipelines.
Image Tiers: Image, Image Quality, and Image Pro
The Grok Imagine image family consists of three model tiers, all powered by Aurora:
Grok Imagine Image: The standard tier. Text-to-image generation at up to 1K resolution (1024×1024), with 7 aspect ratios. The practical starting point for most generation workflows.
Grok Imagine Image Quality: The premium tier. Outputs at up to 2K resolution (2048×2048), with 14 aspect ratios including exclusive ultrawide 20:9 and ultratall 9:20 formats that produce genuine panoramic and full-length compositions, not cropped versions of square outputs. Built on Aurora's autoregressive Mixture-of-Experts network with the largest parameter allocation in the image family.
Grok Imagine Image Pro: Extended capabilities for production pipelines, with up to 10 outputs per request.
Image Output Specifications
| Specification | Grok Imagine Image | Grok Imagine Image Quality |
|---|---|---|
| Max resolution | 1K (1024×1024) | 2K (2048×2048) |
| Output formats | JPEG · PNG · WebP | JPEG · PNG · WebP |
| Alpha channel | PNG only | PNG · WebP |
| Aspect ratios | 7 | 14 (incl. 20:9 · 9:20) |
| Max outputs/request | 4 | 10 |
| Reference images | Up to 3 | Up to 3 |
| Generation speed | 3–5 seconds | 3–5 seconds |
| Architecture | Aurora (autoregressive MoE) | Aurora (autoregressive MoE) |
Supported aspect ratios (Image Quality tier): 1:1 · 16:9 · 9:16 · 4:3 · 3:4 · 3:2 · 2:3 · 2:1 · 1:2 · 19.5:9 · 9:19.5 · 20:9 · 9:20 · auto
When to Use Grok Imagine for Images
Text rendering is critical. For any content where legible text, logos, or multilingual copy must appear inside the generated image, Grok Imagine Image Quality is the strongest option in the current market. The autoregressive architecture's sequential token generation produces a structural advantage on this task that diffusion models don't reliably match.
Large-format or panoramic output. The exclusive 20:9 and 9:20 aspect ratios in Image Quality produce genuine wide-format compositions suited to banner advertising, cinematic establishing shots, and full-length character visualization.
Editing workflows. The natural language editing capability, particularly background replacement and style transfer on existing images, is well-suited to brand asset production where a base image needs to be adapted across multiple contexts.
Integrated image-to-video pipeline. Generated images feed directly into Grok Imagine's video generation. If your workflow involves generating a scene as a still image and then animating it, both steps happen within the same model family with consistent visual identity maintained from image to video.
Part 2: Video Generation (Grok Imagine 1.0)
What Changed in 1.0
Grok Imagine's video generation has been in staged rollout since August 2025, when it first appeared for X Premium subscribers on iOS. The 1.0 release on February 2, 2026 was the first version to reach general API availability and represented a meaningful capability upgrade from the earlier preview:
- Clip duration extended from 6 seconds to 10 seconds
- Resolution upgraded to 720p (from 480p in preview)
- Audio quality significantly improved: dialogue, ambient sound, and sound effects are considerably more natural in 1.0 than in the preview version
- API access opened beyond X (Twitter) for the first time, available to developers through xAI's API and third-party platforms
At the time of the 1.0 release, xAI reported that Grok Imagine had generated 1.245 billion videos in the prior 30 days, reflecting the scale of usage built up during the X-native preview period.
Video Architecture: Why I2V Is the Strongest Mode
Grok Imagine's video generation runs on the same Aurora autoregressive architecture as its image generation, and this has a direct consequence for image-to-video quality: the source image is treated as the literal first frame, not as a visual reference or style guide.
Most video models accept reference images as conditioning signals. The model learns from them and produces output informed by them, but doesn't strictly preserve them. Aurora generates from the source image outward, with the first frame being the source image itself. The model then predicts subsequent frames sequentially from that anchor.
In practice, this gives Grok Imagine 1.0 the strongest subject identity preservation of any mode available on the platform. The person or object in the source image is the same person or object in the output, with the same facial geometry, same proportions, and same visual identity, because the generation is literally continuing from that specific frame, not approximating it.
For portrait animation, product animation, and any use case where "the subject in the output must match the subject in the input" is a primary quality criterion, this architectural characteristic makes I2V the right starting point.
Text-to-Video
Text-to-video in Grok Imagine 1.0 generates clips 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.
Three creative modes are available:
- Normal: Standard generation for most use cases
- Fun: Stylized, heightened, or exaggerated visual treatment
- Spicy: Mature content mode with content policy applied per platform
The model accepts prompts up to 5,000 tokens (context window), providing substantial space for detailed scene direction, camera movement specification, character description, and audio cues.
Prompt structure for T2V:
Grok Imagine responds well to prompts that specify: 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 the wind. She turns slowly toward camera. Golden light from the west, lens flare on the turn. Wave sounds, distant seagulls, quiet wind."
Image-to-Video
I2V takes a still image as input and generates a video clip with the source image as the first frame. The model accepts image URLs as direct input and generates the continuation of that visual state, including movement, atmosphere, and audio, based on the text prompt.
Key I2V capabilities in Grok Imagine 1.0:
Hard first-frame anchoring. The source image is not a style reference; it is frame 1. Subject identity, composition, lighting, and color are carried directly into the generation from the input image.
Motion prompt control. The text prompt specifies what happens in the frames after the first: camera movement, subject action, atmospheric changes, and audio. The model applies these to the anchored visual state.
Native audio. Audio is generated alongside the video in the same pass, with dialogue, ambient sound, sound effects, and background music driven by the visual content and text prompt.
Aspect ratio flexibility. The output aspect ratio can be specified independently of the input image's aspect ratio across 7 supported formats.
Video Output Specifications
| Specification | Details |
|---|---|
| Model version | Grok Imagine 1.0 |
| 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 (I2V) | Up to 7 |
| Context window | 5,000 tokens |
| Generation speed | ~30 seconds |
| Training compute | 110,000 NVIDIA GB200 GPUs (Colossus) |
Source: xAI Imagine API · Grok Imagine API · as of June 2026
The Aurora Architecture: Why It Matters
Most AI image and video models available in 2026 are built on diffusion architectures. Diffusion works by starting from a random noise field and progressively denoising toward a coherent output through many iterative steps.
Aurora works differently. It is an autoregressive model: it generates output tokens sequentially, predicting each token based on all previous tokens. The model builds the image or video from left to right, top to bottom, or in the case of video, frame by frame, rather than refining a noisy whole.
What this means in practice:
Stronger compositional reasoning. Because Aurora generates in sequence rather than in parallel, it maintains awareness of what has already been generated when predicting what comes next. This produces tighter spatial logic, so elements in the image are more consistently positioned relative to each other and objects behave more coherently in video.
Better text understanding within images. The sequential generation process naturally handles the semantic relationship between text content and its visual placement. This is why Grok Imagine's text rendering outperforms diffusion models: Aurora doesn't have to figure out where a word goes and what it means separately. Both are part of the same sequential prediction.
Harder first-frame anchoring in I2V. When the source image is provided as the starting sequence of tokens, the autoregressive model continues that specific sequence rather than conditioning on it loosely. The first frame is not a soft reference; it is literally where the generation begins.
Trade-offs. The sequential generation process is computationally intensive and tends to be slower than single-pass diffusion for comparable output resolutions. Grok Imagine's ~30-second video generation time reflects this architecture. The quality characteristics that Aurora's sequential reasoning enables are the trade-off for the speed advantage that some diffusion models offer.
Real-World Use Cases
Portrait animation for content creators and social media Upload a portrait photo and prompt for natural head movement, eye contact with camera, and ambient audio. The I2V hard-anchor architecture ensures the person in the photo is the person in the video. Short 6–10 second format is native to the content types that benefit most from animated portraits: profile animations, story content, speaking clips.
Product photography animation Animate product still photographs into video with controlled camera movement and synchronized ambient sound. The source image as frame 1 ensures product identity, including color, shape, and branding, is preserved exactly. Combine with the image generation pipeline: generate a product shot with Grok Imagine Image Quality, then animate with I2V.
Brand and marketing asset production Use image generation for initial asset creation (Aurora's text rendering handles brand names and taglines accurately), then animate selected assets into short video formats. That keeps visual consistency across both steps.
Concept art and pre-production visualization Generate visual concepts at 2K resolution across wide format ratios, then animate key frames into short motion studies. The 20:9 panoramic ratio in Image Quality tier is particularly suited to establishing shots and environment visualization.
Social-first short-form content Native 9:16 aspect ratio support, sub-30-second generation time, and integrated audio make Grok Imagine well-suited for high-volume social content production. The three creative modes, Normal, Fun, and Spicy, make it easy to adjust style without prompt rewriting.
How Grok Imagine 1.0 Compares to Alternatives
vs. Seedance 2.0: Seedance 2.0 offers significantly richer multimodal reference inputs (up to 9 images + 3 video + 3 audio simultaneously), stronger physical realism in motion, and multi-shot native generation. Grok Imagine 1.0's advantage is in subject identity preservation via hard first-frame anchoring and in image generation quality, particularly text rendering. For reference-heavy production work, Seedance 2.0's input surface is broader. If identity fidelity is the primary criterion, Grok Imagine 1.0's architecture is more naturally suited for portrait and product I2V.
vs. Veo 3.1: Veo 3.1 leads on audio synchronization precision, 4K resolution, and Scene Extension for longer-form content. Grok Imagine 1.0's generation speed (~30 seconds) is faster than Veo 3.1's Quality tier, and the Aurora image generation provides a stronger image pipeline for creators who work across both still and video formats within the same workflow.
vs. Kling 3.0: Kling 3.0 offers significantly longer single-pass clip generation (up to several minutes) and native 4K output. Grok Imagine 1.0's hard first-frame anchor in I2V and the Aurora image generation integration represent distinct workflow advantages for use cases where image-to-video continuity and image generation quality are the primary criteria.
Known Limitations in Grok Imagine 1.0
10-second maximum clip duration. For content beyond 10 seconds, clips must be assembled from separate generations. There is no native Scene Extension equivalent in 1.0.
720p maximum video resolution. For content requiring 1080p or 4K final output, Grok Imagine 1.0 is not the appropriate choice. Image generation reaches 2K; video stays at 720p in this version.
~30-second generation time. The autoregressive architecture's sequential process takes longer than some diffusion-based models at equivalent quality settings. For workflows requiring high-volume rapid generation, this affects throughput.
Anatomy and hands. Early users identified anatomical inaccuracies, particularly with hands, in image generation. This is a known limitation of the Aurora architecture that xAI has been progressively addressing across versions. In 1.0, this is less prominent than in earlier releases but remains a known area where results may require regeneration.
Multi-person lip sync. Like most current models, Grok Imagine 1.0 produces its most reliable lip sync results with single speaking characters. Multi-person simultaneous dialogue scenes are more variable.
Access through X ecosystem. Grok Imagine is integrated into X (Twitter) and the xAI ecosystem. API access for third-party platforms, including our own, is available but follows xAI's API terms and rate limits.
For current production work, Grok Imagine 1.0 is the version to evaluate today.
Frequently Asked Questions
What is the difference between Grok Imagine image and video generation? They are separate modes within the same model family. Image generation uses Aurora to produce still images at up to 2K resolution. Video generation (Grok Imagine 1.0) uses Aurora to produce 720p video clips of 6–10 seconds with native audio. Both are available on our platform. Image generation is faster (3–5 seconds); video generation takes approximately 30 seconds.
What makes Aurora different from diffusion models? Aurora is an autoregressive model that generates output sequentially, token by token, rather than starting from noise and denoising progressively. This gives it stronger text rendering, tighter compositional reasoning, and harder first-frame anchoring in image-to-video. The trade-off is speed: sequential generation is computationally intensive, which contributes to Grok Imagine's ~30-second video generation time.
Why does Grok Imagine preserve subject identity so well in I2V? Because the source image is not a reference. It is literally the first frame of the generation. Aurora continues the token sequence starting from the source image, so the subject's visual identity is built into the generation state from frame 1. It is not approximated from a description of the source image.
Does Grok Imagine generate audio automatically? Yes. Audio is generated natively in the same pass as the video, with dialogue, ambient sound, sound effects, and background music all produced alongside the visual content. There is no separate audio generation step.
What is "Spicy mode"? Spicy is one of three creative modes available in Grok Imagine video generation (alongside Normal and Fun). It enables mature content generation. Availability of Spicy mode on third-party platforms, including ours, depends on platform content policy. Check our platform's content guidelines for current availability.
What happened in January 2026 with Grok Imagine? In early January 2026, Grok Imagine faced significant backlash following widespread misuse for generating non-consensual deepfake imagery through X's image editing features. xAI restricted free-tier access to image generation and editing starting January 9, 2026, requiring paid X subscriber status for access through X. API access through partner platforms was not affected by this restriction.
Grok Imagine 1.0, covering both image generation and video generation, is available on our platform today.
