xAI's Grok Imagine Video 1.5 Hits #2 With Native Audio in One Pass
xAI's grok-imagine-video-1.5-preview lands #2 on the image-to-video leaderboard with native synchronized audio baked into every generation
Takeaways
- xAI launched grok-imagine-video-1.5-preview, ranking #2 on the Artificial Analysis Image-to-Video (With Audio) leaderboard with Elo 1,110.
- Native synchronized audio is generated in the same inference pass as video, including sound effects, dialogue, and ambient music, no separate step needed.
- Priced at $8.40/min at 720p ($0.14/sec), with audio bundled; available now via the xAI API at
grok-imagine-video-1.5-preview. - Image-to-video only in this preview; clips up to 15s at 720p/24fps, with text-to-video handled by a separate endpoint.
- Key limitation: video quality degrades after 2-3 chained extensions; 1080p output and multi-extension stability are on the roadmap.
- ByteDance's Seedance 2.0 still leads the leaderboard (Elo 1,195) with 1080p output and up to 12 mixed reference inputs per generation.
xAI just dropped grok-imagine-video-1.5-preview, a new image-to-video model that immediately claimed a top-tier spot on the Artificial Analysis Video Leaderboard. The model sits at #2 in the Image-to-Video (With Audio) rankings, behind only ByteDance's Seedance 2.0, and holds #3 in the no-audio rankings. For a preview-stage release, that's a strong opening statement in an increasingly crowded field.
What it actually does
The model takes a single still image and turns it into fluid, cinematic video. Give it a starting frame and a prompt describing the motion, and it animates the scene, including camera moves, atmosphere, and physics, while staying faithful to your source image. That fidelity to the source is the key design choice here: the model holds detail and lighting from the input frame, so the result continues the original image rather than reinterpreting it.
The headline capability that separates 1.5 from its predecessor is native audio. You feed it a still image plus a motion-focused prompt, and it produces a short clip with natively generated, synchronized audio, including dialogue, sound effects, ambient sound, and music, created in the same inference pass as the video, rather than bolted on afterward. That single-pass audio is a meaningful differentiator: silent AI clips increasingly feel unfinished. A product turntable needs subtle room tone or mechanical sound. A character animation needs breath, cloth movement, footsteps, or environmental ambience. A cinematic shot needs sound design that matches the visual mood.
The leaderboard picture
According to the live Artificial Analysis Image-to-Video leaderboard, the current rankings with audio look like this:
| Rank | Model | Elo | Price / min |
|---|---|---|---|
| 1 | Dreamina Seedance 2.0 720p | 1,195 | $9.07 |
| 2 | grok-imagine-video-1.5-preview | 1,110 | $8.40 |
| 3 | HappyHorse-1.0 (Alibaba-ATH) | 1,093 | $13.20 |
| 4 | Veo 3.1 (Google) | 1,089 | $24.00 |
| 5 | grok-imagine-video (xAI) | 1,080 | $4.20 |
The Elo score here is derived from blind human preference votes, not automated benchmarks, which makes it harder to game. According to the latest test results from Arena.ai, Grok Imagine Video 1.5 demonstrates a significant improvement of +52 Elo points compared to the current 1.0 model. The gap to Seedance 2.0 is real though: an 85-point Elo difference is statistically significant at the sample sizes involved.
Where it shines, where it doesn't
The model is built for a specific workflow and it does that workflow well:
- Complex, multi-element directions land. Shot type, camera move, and timing cues written into the prompt show up in the finished clip.
- The reference-to-video workflow accepts a reference image to maintain visual consistency throughout a generated video, keeping a specific character's appearance, a product's design, or a location's visual style stable across the entire clip. This makes it well-suited for character-driven content and brand-consistent video production.
- The model also works well for sequences. Stage each frame, animate it, and chain the shots together into longer scenes that keep a consistent look across an entire project.
The limitations are worth knowing before you build on it:
- It is weaker for long continuous scenes, complex multi-character choreography, and exact frame-by-frame control.
- Community testing confirms that video quality degrades visibly after 2 to 3 chained extensions. xAI has not confirmed a fix timeline. Most creators currently export individual clips and combine them in a separate video editor for longer productions.
- Clips top out at 720p , while Seedance 2.0 outputs at 1080p natively. Higher resolutions are on the roadmap but not yet available.
-
This preview model currently does not support text-to-video. Every request needs an input image.
There is a separate
grok-imagine-videoendpoint for text-to-video.
How it's built
The model runs on xAI's Aurora engine, an autoregressive mixture-of-experts architecture that predicts tokens across interleaved text, image, video, and audio modalities. Autoregressive here means the model generates the output one chunk at a time, each step conditioned on everything before it, which is how it can keep audio and video in tight sync without a separate alignment step. The video pipeline also integrates technology from Hotshot, a video generation startup xAI acquired.
Pricing and access
xAI lists the preview at $0.08 per second at 480p and $0.14 per second at 720p, with a $0.01 image input cost. That works out to $4.80/min at 480p and $8.40/min at 720p. Audio is bundled into that rate, which matters for the real cost-per-deliverable calculation: you're not paying separately for a sound design pass.
Input formats supported include JPG, JPEG, PNG, WEBP, GIF, and AVIF. Output is H.264 MP4 at 24fps across seven aspect ratios, at 480p or 720p resolution.
Clips can run up to 15 seconds.
The model is available now via api.x.ai, identified by the alias grok-imagine-video-1.5-2026-05-30. A broader consumer rollout to X Premium tiers is still in progress.
Getting started takes a few lines of code:
import os
import xai_sdk
client = xai_sdk.Client(api_key=os.getenv("XAI_API_KEY"))
response = client.video.generate(
prompt="Slow cinematic push-in as embers drift across the battlefield",
model="grok-imagine-video-1.5-preview",
image_url="https://your-host.com/image.jpg",
duration=10,
resolution="720p",
)
print(response.url)The competitive context
The image-to-video space has gotten genuinely competitive fast. xAI is now competing directly with video AI providers like Seedance and Google's Veo. Seedance 2.0 is the current leader and it has a meaningful edge: it accepts up to 9 images, 3 video clips, and 3 audio files in a single generation , and outputs at 1080p. Grok 1.5 is more focused, accepting a single image and a prompt, but it's also simpler to integrate and priced competitively against the field.
The practical use cases where this model makes the most sense right now:
- Product animation: Turn a product photo into a motion clip with ambient sound, without a separate audio workflow.
- Social content at scale: The model outputs at 720p, 24fps, sufficient for professional publishing on X, TikTok, Instagram Reels, and YouTube Shorts.
- Storyboarding and pre-viz: xAI's launch post specifically highlights staging frames and chaining shots together, a useful workflow for directors, animators, and agencies: create a set of still frames, animate each one, then cut them together into a longer scene.
- Character animation: If you have a character illustration or AI-generated portrait, the model can add expression, camera motion, and atmosphere while preserving the base design.
The broader signal here is that native audio in video generation is quickly becoming table stakes. Grok 1.5 joins Seedance 2.0 and Veo 3 in treating audio as a first-class output rather than a post-processing step. Models that still separate the two pipelines are starting to look like a generation behind.