Grok Imagine 1.5 reached #1 on the I2V Arena in June 2026 (+52 Elo). Covers Video Extend, Fast tier, grok image to video specs, and migration from 1.0.
Grok Imagine 1.5: AI Image to Video — 2026 Model Guide
Developer: xAI · Architecture: Aurora (Autoregressive MoE) · API Preview: June 3, 2026 · GA Released: June 16, 2026 · API model ID: grok-imagine-video-1.5
Version note: This page covers Grok Imagine 1.5, xAI's current recommended video generation model as of July 2026. For the complete model family guide including Aurora architecture and image generation, see the Grok Imagine complete guide. For the previous version's technical reference, see Grok Imagine 1.0.
What Changed in Grok Imagine 1.5
Grok Imagine 1.5 moved from API preview (grok-imagine-video-1.5-preview, June 3, 2026) to general availability (grok-imagine-video-1.5, June 16, 2026) as xAI's most significant upgrade to the video generation product since its initial public release. The changes are concrete and measurable.
Extended clip duration. Maximum single-pass generation increased from 10 seconds to 15 seconds, a 50% increase. Longer single-pass clips reduce the number of Extend from Frame calls required for content in the 10–15 second range and simplify production pipelines for most short-form content.
1080p resolution support. Version 1.5 added 1080p as a video output resolution option alongside the existing 480p and 720p. This is a meaningful addition for workflows targeting broadcast-quality or high-resolution platform delivery, and brings Grok Imagine 1.5 into the resolution range used by professional video production tools.
Fast speed tier. Version 1.5 introduced a Fast tier that reduces generation latency by approximately 40%. Per xAI's announcement: "Video 1.5 Fast generates 6-second, 720p clips in approximately 25 seconds — nearly twice as fast as the previous model." The Fast tier is available through the grok.com/imagine interface and X mobile apps; the Standard API tier is available for production pipelines requiring the highest quality output.
Quality improvement reflected in Arena ranking. The +52 Elo point gain on the Artificial Analysis Image-to-Video Arena is the measured evidence of 1.5's quality improvement. In Arena comparisons, evaluators assess overall video quality in a blind pairwise format; in practice, that preference signal typically reflects motion naturalness, temporal coherence, prompt adherence, and audio-visual synchronization. The +52 Elo gain indicates 1.5 produced systematically preferred outputs across those dimensions.
Audio synchronization improvements. Lip-sync accuracy, audio-visual timing precision, and the naturalness of generated dialogue and ambient sound improved substantially in 1.5 relative to 1.0.
Video Extend quality improvements. The Extend from Frame feature, originally introduced on March 2, 2026 as part of the 1.0-era model, received quality improvements in 1.5: better scene continuity across clip boundaries, stronger character consistency, and more reliable audio coherence when extending.
Why Grok Imagine 1.5 Ranked #1
The Artificial Analysis Image-to-Video Arena is an independent benchmark that uses blind pairwise comparison to generate Elo ratings for video generation models. Human evaluators select the better video in head-to-head comparisons without knowing which model produced each clip. Elo scores are calculated from thousands of such comparisons, making the ranking a direct measure of human preference rather than a task-specific benchmark.
Grok Imagine 1.5 entered the Arena as a preview on May 31, 2026 and subsequently reached the #1 position on the I2V leaderboard. The model's Elo score peaked at approximately 1,466 in a June 23, 2026 snapshot, a +52 Elo improvement over version 1.0. At that position, it surpassed:
- Seedance 2.0 (ByteDance): previously recognized for strong multimodal reference input capacity
- HappyHorse 1.0: strong photorealistic motion physics
- Google Veo 3.1: Google's flagship 4K video model with Scene Extension
- Kling 3.0 (Kuaishou): long-form generation leader
What the #1 ranking measures, and what it does not:
The Arena ranking reflects human preference for overall video quality in blind evaluation: motion realism, subject identity preservation, temporal coherence, prompt adherence, and audio-visual synchronization within the operational constraints of Grok Imagine 1.5's design (clips up to 15 seconds at up to 1080p).
The ranking does not reflect resolution capability above 1080p (Veo 3.1 and Kling 3.0 generate at 4K), clip duration beyond 30 seconds with Video Extend, or multimodal input depth. For use cases where those factors are the primary decision criteria, the Arena ranking should be considered alongside those structural constraints.
Elo scores fluctuate continuously as new comparisons are submitted. The ~1,466 figure reflects a June 23, 2026 snapshot; current rankings should be verified directly at artificialanalysis.ai.
Grok Image to Video in Grok Imagine 1.5
Image-to-video is architecturally Grok Imagine's strongest mode, and the 1.5 release brought the most significant I2V quality improvement since the product launched. The model earned its #1 Arena ranking primarily through I2V performance: motion naturalness, subject identity preservation, and audio-visual coherence in animated-from-image clips.
How I2V Works in 1.5
Provide a still image as input alongside a text prompt specifying motion, camera behavior, and audio. Aurora treats the source image as the literal first frame, not a style reference or conditioning signal, but the actual starting point of the token sequence from which all subsequent frames are generated. Subject identity, facial geometry, lighting, and composition carry directly from the source image into the video because the model is continuing from that specific token sequence, not predicting a new one.
Core I2V Capabilities in Version 1.5
- 15-second maximum single-pass duration at 24 FPS (up from 10 seconds in 1.0)
- Resolution options: 480p, 720p, 1080p (specify output resolution independently of input image dimensions)
- Aspect ratio flexibility: output aspect ratio specifiable across 7 formats independently of input image dimensions
- Native audio: dialogue, ambient sound, SFX, and BGM generated in the same pass as the video
- Improved lip sync: single-character audio-visual synchronization improved over 1.0
When I2V Is the Right Choice vs. T2V
I2V is architecturally stronger in Grok Imagine 1.5 because it benefits from Aurora's hard first-frame anchoring. For any workflow where subject identity must match a specific visual reference (portrait animation, product animation, brand asset animation, character continuity across a series), I2V is the correct mode.
T2V is appropriate for conceptual content where no source image exists. However, T2V produces weaker subject identity coherence and compositional consistency than I2V because the model generates all frames from the text description without a visual anchor point. If subject identity matters, generate a reference image first (Grok Imagine Image Quality is the most cohesive choice) and use I2V.
Text-to-Video in Grok Imagine 1.5
Text-to-video generates clips from natural language prompts without a source image. The model produces video at 24 FPS with native audio in clip durations from 6 to 15 seconds.
T2V in 1.5 benefits from the same quality improvements as I2V: better motion physics, sharper detail, improved temporal coherence, and enhanced audio-visual synchronization. The 15-second maximum also extends T2V utility for short-form content that previously required chaining two 6-second clips.
Three creative modes: Normal (standard generation), Fun (stylized treatment), Spicy (mature content, availability subject to platform policy).
For prompt structure, motion control, and T2V best practices, the same techniques from 1.0 apply. Scene description, subject and action, camera behavior, atmosphere, and audio direction form the most effective prompt structure.
Video Extend
The Extend from Frame feature allows chaining video clips to produce longer sequences beyond the single-pass 15-second limit. The feature was originally introduced on March 2, 2026 as part of the 1.0-era model; version 1.5 improved scene continuity, character consistency across clip boundaries, and audio coherence when extending.
How Video Extend Works
Generate an initial clip (6–15 seconds). Then call the Extend from Frame endpoint with the result, providing a continuation prompt describing the next segment's action, camera, and audio. Aurora treats the final frames of the input clip as the anchor and generates the next segment, maintaining visual continuity, subject identity, and scene context.
Extension parameters:
- Input: Previous video clip URL or file reference
- Continuation prompt: Text describing the next segment's action, camera movement, and audio
- Extension duration: 2–10 seconds per extension call
- Maximum total duration: Approximately 30 seconds through chaining; quality consistency may decrease with successive extensions
Practical Workflows with Video Extend
Extended portrait speaking clip: Generate a 12-second portrait animation anchored from a source headshot, then extend with a continuation prompt for an additional 10–12 seconds of speaking. The hard first-frame anchor maintains consistent facial geometry across the full extended output.
Multi-shot narrative: Build a two-shot sequence by generating a 10-second establishing shot, then extending with a different camera angle or subject action in the continuation. Total: up to 20 seconds of coherent narrative from two chained generations.
Product reveal with follow-up: Generate a 10-second product reveal, then extend with a 10-second product-in-use scene. The continuation prompt can direct a natural scene cut or a continuous camera movement.
Known constraints: Video Extend produces best results when the continuation prompt describes a logical progression from the final state of the input clip. Abrupt scene cuts, major subject changes, or drastic lighting shifts produce weaker continuity. Natural extensions (continuing motion already in progress, maintaining a scene's established logic) are more reliable than conceptual jumps.
Speed Tiers: Fast vs. Standard
Version 1.5 introduced two distinct speed tiers with different quality-latency trade-offs:
Grok Imagine Video 1.5 Fast
- Generation speed: ~25 seconds for a 6-second 720p clip (xAI: "nearly twice as fast as the previous model")
- Access: grok.com/imagine interface and X iOS/Android apps
- Best for: High-volume workflows, social content production, rapid iteration and candidate selection
- Quality: High quality relative to most competing models; measurably below Standard tier in direct comparison
Grok Imagine Video 1.5 (Standard API)
- Model ID:
grok-imagine-video-1.5 - Access: xAI Imagine API
- Best for: Final deliverables, production pipelines, content where quality is the primary criterion
- Quality: Full quality output corresponding to the #1 I2V Arena benchmark results
Both tiers use Aurora's autoregressive architecture and maintain hard first-frame anchoring in I2V mode. The selection between Fast and Standard is a quality-speed trade-off within the same model family, not a choice between different architectural approaches.
Output Specifications
| Specification | Grok Imagine Video 1.5 |
|---|---|
| Architecture | Aurora (Autoregressive MoE) |
| Clip duration (single-pass) | 6–15 seconds |
| Total duration (Video Extend) | ~30 seconds |
| Resolution | 480p · 720p · 1080p |
| 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; improved in 1.5) |
| Creative modes | Normal · Fun · Spicy |
| Context window | 5,000 tokens |
| I2V Arena Elo (peak, Jun 23, 2026) | ~1,466 (#1 ranked) |
| Generation speed (Fast tier) | ~25 seconds for a 6-second 720p clip |
Source: xAI Grok Imagine Video 1.5 Announcement · Artificial Analysis Image-to-Video Arena · As of July 2026
Access and Availability
X Subscription Access
- Free tier: Limited weekly generation quota shared across image and video
- X Premium ($8/month): Increased quota; access to Grok Imagine 1.5 Fast on grok.com/imagine
- SuperGrok ($30/month): Substantially higher generation limits; access to 1.5 Fast
As of June 2026, X uses a unified weekly credit pool shared across image and video generation. Video generation consumes significantly more credits per request than image generation.
Third-Party Platform Access
Grok Imagine 1.5 is available on VioEvo and other third-party platforms without requiring a direct xAI API account. Check platform documentation for current tier availability and per-generation pricing.
Real-World Workflows with Grok Imagine 1.5
Portrait animation with extended speaking sequence. Generate a 12-second portrait animation anchored from a source headshot, then use Video Extend to continue through a 25-second speaking sequence at 720p. The hard first-frame anchor maintains consistent facial geometry and identity across the full extended output. Native audio handles dialogue without a separate recording step. Output format: 9:16 for direct social publishing.
Product photography to 15-second demo video. Generate a product shot at 2K resolution using Grok Imagine Image Quality for accurate text rendering and branding, then animate with I2V at 1080p for a 15-second product demonstration. Both steps use the same Aurora model family for visual identity continuity from still to motion.
Fast tier candidate selection for social content. Use the Fast tier (~25 seconds per clip) to generate multiple visual directions from the same brief: different camera angles, motion approaches, or audio moods. Select the strongest candidate, then re-generate with the Standard API for the final deliverable. The 9:16 aspect ratio fits Instagram Reels, TikTok, and YouTube Shorts natively.
How Grok Imagine 1.5 Compares
vs. Seedance 2.5: Seedance 2.5 offers richer multimodal reference inputs (supporting multiple video, image, and audio references simultaneously) and stronger physical realism in complex motion. Grok Imagine 1.5's advantages are in subject identity preservation via hard first-frame anchoring and in the Fast tier's generation speed for short-form content. For reference-heavy production work where multiple input types need to inform the output, Seedance 2.5's input surface is broader. For portrait animation and product animation where one strong source image defines the subject, Grok Imagine 1.5's architecture is more naturally suited.
vs. Google Veo 3.1: Veo 3.1 leads on resolution (4K vs. 1080p), Scene Extension for longer narrative content, and audio synchronization precision in multi-person dialogue scenarios. Grok Imagine 1.5's advantages are in generation speed (Fast tier is substantially quicker than Veo 3.1's Quality mode), in the integrated image generation pipeline, and in subject identity fidelity when I2V hard first-frame anchoring is the decisive quality criterion. Veo 3.1 is appropriate when 4K output or extended long-form narrative is required; Grok Imagine 1.5 is appropriate when 1080p is sufficient and speed, I2V subject fidelity, or cost-per-second are primary factors.
vs. Kling 3.0: Kling 3.0 offers substantially longer single-pass clip generation and native 4K output, making it the structural choice for long-form content and broadcast-grade final deliverables. Grok Imagine 1.5's advantages are in subject identity preservation via hard first-frame anchoring and in generation speed for content within the 15-second single-pass range. For workflows under 30 seconds where subject fidelity matters more than resolution or clip duration, Grok Imagine 1.5 provides stronger identity coherence. For multi-minute single-pass generation or 4K final output, Kling 3.0 is more appropriate.
vs. HappyHorse 1.0: HappyHorse 1.0 is recognized for strong photorealistic motion physics and naturalistic camera movement in T2V. Grok Imagine 1.5 surpassed HappyHorse on the I2V Arena leaderboard through improved temporal coherence and subject identity preservation. HappyHorse remains a strong option for T2V workflows where no source image exists and photorealistic motion is the primary criterion. For I2V workflows anchored from a specific source image, Grok Imagine 1.5's architectural approach is structurally stronger.
Known Limitations in Grok Imagine 1.5
1080p is the maximum video resolution. For workflows requiring 4K final output, Veo 3.1 and Kling 3.0 are the current options. The resolution ceiling reflects the computational constraints of Aurora's autoregressive sequential generation at scale.
30-second total duration limit. For content beyond 30 seconds, clips must be assembled from separate Video Extend sequences or joined in post-production. There is no equivalent to Veo 3.1's Scene Extension for continuous longer-form narrative beyond 30 seconds.
T2V produces weaker subject coherence than I2V. Without a source image anchor, T2V generates subject appearance from the text description, producing more variable identity coherence over the clip. For any use case where subject identity must be consistent, generate a source image first and use I2V.
Single-character lip sync most reliable. Multi-person simultaneous dialogue produces more variable lip-sync accuracy across all current models including 1.5. For tight multi-speaker audio-visual synchronization, Veo 3.1's dedicated audio synchronization may produce more consistent results.
Standard API generation speed is slower than some diffusion alternatives. Even with the Fast tier's 40% improvement, Aurora's sequential generation process is computationally intensive. High-volume pipelines with strict throughput requirements should factor generation time per clip into workflow planning.
Reference-to-Video (multi-image input) is not available in 1.5. The @image1…@image7 multi-reference image syntax is supported only by grok-imagine-video (1.0). In Grok Imagine 1.5, image-to-video takes a single source image as the literal first frame. For workflows requiring multiple conditioning images, use the 1.0 endpoint or consider Seedance 2.0/2.5, which offer broader multi-reference input support.
Migrating from Grok Imagine 1.0
For existing workflows using Grok Imagine 1.0 (grok-imagine-video), the migration to 1.5 is designed to be straightforward.
API compatibility. Grok Imagine 1.5 is backward compatible with 1.0 prompts and parameters. Existing prompts, reference image handling, aspect ratio specifications, and creative mode selections require no changes. The migration is a model identifier update: replace grok-imagine-video with grok-imagine-video-1.5 in your API calls.
Prompt structure unchanged. Text prompt syntax, reference image inputs, and creative mode parameters remain consistent between 1.0 and 1.5. Prompts that performed well in 1.0 will produce similar or better results in 1.5 without modification.
New features are additive. The Fast tier, 1080p output, and extended 15-second clip duration are new options that do not affect existing I2V or T2V workflows. Adopt them incrementally as your workflow benefits.
Quality improvement is automatic. The +52 Elo improvement applies to all generation requests on the 1.5 endpoint. No prompt tuning is required to benefit from the quality upgrade.
Pricing change to plan for. Version 1.5 is priced higher than 1.0 at both 480p and 720p. Update budget allocations before migrating high-volume workflows. The 1.5 480p tier can partially offset this in cost-sensitive pipelines where 720p is not required. See xAI API documentation for current rates.
Frequently Asked Questions
Why did Grok Imagine 1.5 rank #1 on the I2V Arena leaderboard? The #1 ranking reflects human evaluators' blind preference for Grok Imagine 1.5's output when compared pairwise against competing models. Evaluators consistently preferred its motion realism, subject identity preservation, temporal coherence, and audio-visual synchronization quality. The ranking measures perceived quality within the model's operational constraints (clips up to 15 seconds at up to 1080p) and does not reflect resolution capability above 1080p or duration beyond 30 seconds with Video Extend.
What is the difference between the Fast and Standard tiers? The Fast tier reduces generation time by approximately 40% through inference optimizations. Per xAI's announcement, 1.5 Fast generates a 6-second 720p clip in approximately 25 seconds. The Standard API tier produces full-quality output corresponding to the #1 Arena ranking. Fast is appropriate for high-volume production and rapid iteration; Standard is appropriate for final deliverables where quality is the primary criterion.
Why does Grok Imagine 1.5 preserve subject identity better than most other models in I2V? Aurora's autoregressive architecture treats the source image as the literal first frame: the token sequence continues from that starting point. Subject identity is built into the generation state from frame 1 because the model is continuing an existing sequence rather than predicting a new one from a description. In T2V, where no source image exists, the model generates all frames from text, producing weaker identity coherence. This is a structural property of the autoregressive architecture. For the full explanation, see the Aurora Architecture section in the Grok Imagine complete guide.
When did Grok Imagine 1.5 reach general availability?
The API preview (grok-imagine-video-1.5-preview) was officially announced June 3, 2026. General availability (grok-imagine-video-1.5) launched June 16, 2026, including the Fast tier rollout to grok.com/imagine and X mobile apps.
What is Video Extend and how is it different in 1.5 vs. 1.0? Video Extend (Extend from Frame) chains video clips by treating the final frames of an existing clip as the anchor for the next segment. The feature was originally introduced on March 2, 2026 for the 1.0-era model. Version 1.5 improved scene continuity across clip boundaries, character consistency, and audio coherence when extending. Extension increments are 2–10 seconds per call; the maximum total sequence is approximately 30 seconds.
Is the migration from 1.0 to 1.5 API-compatible?
Yes. The migration requires only a model identifier change from grok-imagine-video to grok-imagine-video-1.5. Prompt structure, reference image handling, and output parameters are backward compatible. No workflow restructuring is required to benefit from the quality improvement.
Why does Grok Imagine 1.5 support a maximum of 1080p instead of 4K? Aurora's autoregressive sequential generation is computationally intensive, and the current implementation is optimized for up to 1080p output. Diffusion-based models like Veo 3.1 and Kling 3.0 scale to 4K more efficiently because their parallel denoising process does not have the same per-token sequential cost. xAI has not announced a timeline for higher resolution support.
What happened to Sora 2? OpenAI discontinued the standalone Sora 2 application in April 2026. Sora 2 remains accessible through ChatGPT for Plus and Pro subscribers, but the dedicated video generation interface was discontinued. Grok Imagine 1.5, Veo 3.1, and other models have been the primary alternatives for creators previously using the standalone Sora 2 interface.
Grok Imagine 1.5 is available on VioEvo, supporting image-to-video, text-to-video, and Video Extend workflows. No xAI account required.
