Evidence review · 2026

Grok Imagine Video 1.5 review: verified public evidence

This evidence review is based on official records, provider data, independent public tests, and external benchmarks. It does not claim FrameSignal hands-on testing.

Updated Sep 23, 2026Model brief
External evidence summary. A FrameSignal hands-on review has not been completed. The source reports below are also included in the model details.

Current evidence summary

xAI's generally available Grok Imagine Video 1.5 now covers text-to-video, image-to-video, reference-to-video, first/last-frame control, video editing and extension with native generated audio. This page separates the current canonical 15-second xAI limit and native 1080p T2V/I2V surface from provider-specific subsets and Roko's non-official 30-second route. The source-backed use cases currently point to native-audio text, image and reference-to-video workflows, teams needing first/last-frame, edit and extension controls, developers comparing xai direct pricing with provider route billing, current benchmark and preview-to-ga capability tracking.

Public benchmark snapshots

Each result stays attached to its task, date, sample count, and reported uncertainty. Developer backend timings are kept separate from end-to-end wait tests.

Artificial Analysis Image to Video Leaderboard (With Audio) · Sep 22, 2026

1.5

image to video with audioElo1099 Elo · rank 8 · n=4,861
image to video with audiorank8 rank · rank 8 · n=4,861
image to video with audiosamples4861 comparisons · rank 8 · n=4,861
Open benchmark source ↗

External tests and provider examples

These are reports from the named publishers and providers; FrameSignal did not run these generations.

EvoLink · EvoLink Team · reliability A

provider hands on i2v

EvoLink's product-ad I2V test kept the perfume-bottle composition convincing through a restrained orbit and light sweep, while delicate requested audio was limited.

Reported result: Bottle/composition stayed convincing; restrained orbit/light sweep worked, but requested delicate audio was limited.

Prompt: Premium perfume bottle single-shot test; exact prompt preserved at source.

Settings: source image: true · duration seconds: 6 · resolution: 720p · single continuous shot: true

Open source ↗
EvoLink · EvoLink Team · reliability A

provider hands on talking head i2v

EvoLink's talking-head I2V test produced usable short American-English speech, small head motion and room tone; lip-sync and pronunciation still needed review.

Reported result: Short American-English speech, small head motion and room tone were usable; author says lip-sync/pronunciation still need review.

Prompt: Direct-to-camera English talking-head test; exact prompt preserved at source.

Settings: source image: true · duration seconds: 6 · resolution: 720p · single continuous shot: true

Open source ↗
EvoLink · EvoLink Team · reliability A

provider hands on cinematic i2v

EvoLink's desert-courier cinematic I2V test delivered strong mood, with local deformation risk when wind affected scarf, coat and hair.

Reported result: Strong cinematic mood; local deformation risk increased when wind affected scarf, coat and hair.

Prompt: Desert courier cinematic test with wind, sand, lightning and thunder; exact prompt preserved at source.

Settings: source image: true · duration seconds: 6 · resolution: 720p · single continuous shot: true

Open source ↗
Runware · Runware · reliability A

provider documented t2v

Reported metric: raw generation charge — 1.5 USD

Runware's Raku pottery workshop text-to-video example shows a $1.50 raw generation charge and approximately 3m42s end-to-end wait.

Reported result: Provider displays approximately 3m42s end-to-end and raw output.

Prompt: Raku pottery workshop social ad; full prompt preserved at source.

Settings: duration seconds: 6 · dimensions: 1904x1072

Open source ↗
Runware · Runware · reliability A

provider documented first frame

Reported metric: raw generation charge — 0.99 USD

Runware's first-frame-with-voice example shows a $0.99 raw generation charge and approximately 42 seconds end-to-end.

Reported result: Provider displays approximately 42 seconds end-to-end and raw output.

Prompt: Mobile knife sharpening reel; full prompt preserved at source.

Settings: duration seconds: 7 · resolution: 720p · preset voice: rex · first frame: true

Open source ↗
Runware · Runware · reliability A

provider documented r2v

Reported metric: raw generation charge — 0.73 USD

Runware's three-reference logo-loop example shows a $0.73 raw generation charge and approximately 43 seconds end-to-end.

Reported result: Provider displays approximately 43 seconds end-to-end and raw output.

Prompt: Avalanche forecast center logo loop; full prompt preserved at source.

Settings: duration seconds: 5 · resolution: 1280x720 · reference images: 3

Open source ↗
Runware · Runware · reliability A

provider documented dialogue

Reported metric: raw generation charge — 1.4 USD

Runware's multi-speaker dialogue example shows a $1.40 raw generation charge and approximately 55 seconds end-to-end.

Reported result: Provider displays approximately 55 seconds end-to-end and raw output.

Prompt: Oyster hatchery documentary B-roll with two preset voices; full prompt preserved at source.

Settings: duration seconds: 10 · dimensions: 720x1280 · preset voices: ["iris","atlas"]

Open source ↗

What official evidence suggests

Promising areas

  • Current xAI docs cover T2V, I2V, R2V, first/last-frame, editing and extension
  • Native generated audio with up to 7 reference images and 3 preset voices
  • Native 1080p T2V/I2V is distinguished from the 720p R2V and edit ceilings
  • Seven accepted provider-authored examples preserve prompts, settings, costs or timing

Verify before paying

  • Canonical xAI generation remains capped at 15 seconds; Roko's longer route is non-official
  • Provider subsets do not expose the full xAI capability surface
  • Per-second direct rates and fixed per-generation gateway charges are not directly equivalent
  • Current Artificial Analysis snapshot is rank 8 rather than the June launch-era #1 claim