Skip to content
Grok image
English
English简体中文繁體中文DeutschEspañolPortuguês

Grok Imagine Video 1.5

Grok Imagine Video 1.5 is the video model xAI ships in its Imagine API under the name grok-imagine-video-1.5. This page puts the publisher's own documentation next to the settings this site's service returns for the same product, and marks where the two disagree.

Updated:

Grok Imagine Video 1.5

Choose a mode, write the prompt, then read the quoted cost before you generate.

Grok Imagine Video 1.5
AI Video0
5s
2s15s

This model does not support reference video input

Generated content will appear here

Source material

Official material for this model

Three items the publisher released that could be retrieved and verified, with the publisher's own prompt text where the publisher published one.

The gallery is short on purpose. The publisher released far more footage than this, but those files are not retrievable from this machine, and the poster frames of three of them turned out to be product interface screenshots rather than model output. What is not here is listed in the build record instead of being substituted with something else.

A lone figure standing on a dark plain under a dense field of stars, the Milky Way rising as a bright band across the frame Read the prompt ↗

01 / Official input still

The still xAI's image-to-video example animates

xAI documentation ↗

Prompt published with this item

Generate a slow and serene time-lapse

Open this prompt in the workspace ↗

Prompts are reproduced in the publisher's own English. Where a line is an instruction aimed at the pictured clip rather than the prompt that produced it, the caption says so.

A crested Spartan helmet standing on a battlefield at sunset, smoke and distant figures behind it Read the prompt ↗

02 / Frame from an official trailer

Key frame from the trailer the publisher credits to this model

xAI news ↗

Prompt published with this item

Slow cinematic push-in as embers drift across the battlefield and the helmet's crest stirs in the wind

Open this prompt in the workspace ↗

Prompts are reproduced in the publisher's own English. Where a line is an instruction aimed at the pictured clip rather than the prompt that produced it, the caption says so.

A woman in a pale dress on a balcony over a bay at golden hour, one hand lifted in a small gesture Read the prompt ↗

03 / Frame from an official sample clip

A moment from the clip the publisher edits in its own example

xAI documentation ↗

Prompt published with this item

Give the woman a silver necklace

Open this prompt in the workspace ↗

Prompts are reproduced in the publisher's own English. Where a line is an instruction aimed at the pictured clip rather than the prompt that produced it, the caption says so.

What Grok Imagine Video 1.5 is, in the publisher's own words

Two xAI announcements describe this model. The first, dated Jun 16, 2026 on the publisher's news page, says Grok Imagine Video 1.5 is generally available on the Imagine API, calls it one of "our best image-to-video models yet: better motion, better physics, better audio, at the fastest speeds", and gives the three headings the page is built around: audio and speech, motion and physics, and speed. The second announcement adds image and voice references and native 1080p, and opens by calling it "our best video model, now with text, image, and voice references - generating up to 1080p."

The model identifier is grok-imagine-video-1.5, version 1.0. The publisher lists two aliases, grok-imagine-video-1.5-preview and grok-imagine-video-1.5-2026-05-30, which is why older write-ups carry a May date and an "xAI released" line that does not match the generally available announcement.

Everything in this section is quoted from the publisher, and the sources are archived alongside this page. Nothing on this page is a statement by xAI, and this page has not been reviewed by xAI.

Quotations are left in the publisher's English. Translating a specification is how numbers drift.

What this site's service publishes for the same product

The workspace does not read the publisher's documentation. It reads this site's own service catalogue, which answers with one product entry per model. For Grok Imagine Video 1.5 that entry is available and callable, sits in the family the catalogue calls xAI Grok video, and declares three task types: text to video, image to video, and reference to video. Each of the three carries the same published floor of fourteen points per task and the same ceiling of forty-five.

The catalogue declares reference to video for this product, and the workspace only offers a family in that picker when the mode it receives can carry a video input. This entry takes images, not clips, so the mode is declared by the service but does not appear in the reference-to-video menu.

ModeWhat you supplyReference images
Text to videoa written scenenone
Image to videoa prompt and a still to animate1 to 7
Reference to videoa prompt and visual references1 to 7, but see the note below

The four controls the workspace actually renders

The service publishes a form schema for each mode, and the workspace builds its controls from that schema rather than from the model tables shipped in this site's source code. For this product the schema is prompt, resolution, aspect ratio and duration, with an image field added on the two image-driven modes. The values below are therefore the values a visitor sees, not the values a local defaults file happens to contain.

ControlTypeValues published by the service
prompttext, requiredfree text
resolutionselect480p, 720p - default 480p
aspect_ratioselect2:3, 3:2, 1:1, 9:16, 16:9 - default 16:9
durationselect6, 8, 9, 10, 12 to 30 - default 6

Duration: the two published ranges are not the same range

This is the single most useful thing to know before spending points here. The publisher's API documentation says the duration parameter accepts 1 to 15 seconds. The service this site sells through publishes twenty-three duration options that run from 6 seconds to 30 seconds. Those are different sets with different ends, and both are published by a source you can read for yourself.

The workspace follows the middle row, because the service publishes a duration field and a published field wins over a local default. The third row is kept in this table only so that the difference is visible; nothing in it reaches the form.

SourceRange publishedDefault
xAI API documentation1 to 15 secondsnot stated
This site's service catalogue6, 8, 9, 10, then 12 to 306 seconds
Local fallback table in this site's source2 to 15 seconds, step 15 seconds

Resolution: three tiers upstream, two tiers in this workspace

The publisher documents three output resolutions and prices them per second of generated video: 480p at $0.08, 720p at $0.14 and 1080p at $0.25. It also states that 1080p is supported for text-to-video and image-to-video on this model, and that reference-to-video is capped at 720p. The service this site sells through publishes two of those three tiers, 480p and 720p, and defaults to the cheaper one.

It is worth being exact about the third source. A local fallback table in this site's source lists 720p and 1080p for this family, which is neither the upstream three-tier list nor the two tiers the service publishes. That table only applies when the service publishes no resolution field at all, and the service does publish one, so the form shows 480p and 720p.

The publisher also charges for inputs: an image input is billed, a text prompt is not, and preset-voice audio input is free. Those are the publisher's own API prices, per second, and they are not what this site charges. What this site charges is quoted by the server in points before a task starts.

Sound arrives with the picture, and this workspace has no switch for it

The publisher states that generated videos include an audio track by default, and that passing generate_audio set to false requests a silent video. On the reference-to-video path it goes further: up to three preset voices can be attached per request, addressed in the prompt as index tags, drawn from the same voice roster as the publisher's text-to-speech model, with your own audio files reserved for trusted partners on request.

The service this site sells through publishes no audio field for this product in any of its three modes. So on this page there is no "include audio" toggle and no voice picker, and the fact that the announcement's audio demos exist does not mean a visitor here can direct them.

Read together, the two statements are not a contradiction: the upstream API can carry audio control, and this site's catalogue has not published a field for it. The honest description of the workspace is the second one.

Nothing in this section is a claim that audio is off. It is a statement that this workspace gives you no control over it.

References, and how text-to-video is built underneath

The publisher's reference-to-video documentation sets a maximum of seven reference images per request, and the second announcement repeats that figure as "up to seven references per generation", describing what each one is for: a face, a product, a location. Reference images guide the shot without locking its first frame, which is what separates the mode from image-to-video.

One implementation detail is easy to miss and explains a lot of confusing results. The publisher states that on this model text-to-video runs as text-to-image followed by image-to-video: a first frame is generated from your prompt, then animated. You still send one request, and the intermediate still is not returned to you.

The same documentation notes that image-to-video output defaults to the aspect ratio of the input image unless you pass an aspect ratio, in which case the image is stretched to the ratio you asked for rather than cropped.

Why this page shows stills instead of the publisher's clips

The publisher released a lot of footage for this model: three sound-design demos and three motion demos, a before-and-after speed pair, three feature walkthroughs and a trailer on the announcement page; two clips on the preview page; and four native-1080p reference clips on the references page. Every one of those files is served from data.x.ai or media.x.ai and every one of them answers with an HTTP 403 error page to a normal request, including requests made from a real browser session carrying the publisher's own cookies and referrer. Opened in the publisher's own page, the same files report a media error rather than playing.

So this page shows what it could actually retrieve and verify: one official clip that is served from the publisher's documentation path and does download, four frames sampled from it, one key frame from the trailer the publisher says was made with this model, and the still that the publisher's own image-to-video example animates.

Every one of those items is captioned with what it is and where it came from. The clips that could not be retrieved are listed in this page's build record rather than being described as if they had been watched.

Three further official images were retrieved and then set aside rather than shown: the poster frames for the Projects, Multiple agents and Search walkthroughs are application screenshots, and a screenshot of a product interface is not evidence of what a model generates.

Examples

The one official clip this page could retrieve: 8.1 seconds, 848 by 480, 24 frames per second, with an audio track, published by xAI as the source video for its video-editing example. It is the input to that example, not an output of Grok Imagine Video 1.5, and it is shown here as published source material rather than as a sample of this model.

xAI documentation, video editing page, retrieved 2026-09-19 from data.x.ai/docs/video-generation/portrait-wave.mp4. · Third-party material. Rights remain with the publisher; shown with source attribution and no claim of ownership.

A woman in a pale dress standing at a balcony railing, hands resting in front of her, the sea and town behind her at low golden light

Frame sampled at 1.0 second. The opening beat: she is still, weight settled, the camera holding on the balcony.

Frame sampled by this site from xAI's published sample clip, retrieved 2026-09-19. · Third-party material. Rights remain with the publisher; shown with source attribution and no claim of ownership.

The same woman on the balcony, turned slightly, hair and dress moving, the harbour lights beginning to read against the sky

Frame sampled at 3.0 seconds. The turn: the subject reads as rotating rather than sliding, which is the part of a clip that motion ratings are usually about.

Frame sampled by this site from xAI's published sample clip, retrieved 2026-09-19. · Third-party material. Rights remain with the publisher; shown with source attribution and no claim of ownership.

The woman raising one hand in a small wave toward the camera, the bay and its pale buildings filling the background

Frame sampled at 5.0 seconds. The gesture beat, and the frame this page uses as the gallery card.

Frame sampled by this site from xAI's published sample clip, retrieved 2026-09-19. · Third-party material. Rights remain with the publisher; shown with source attribution and no claim of ownership.

The woman with both arms lowered, close to the railing, the light on the water behind her deepening

Frame sampled at 7.0 seconds. The close of the clip: the same framing, the subject returned to rest. Worth comparing against the 1.0 second frame, because holding a subject in place across eight seconds is the failure mode this kind of shot exposes first.

Frame sampled by this site from xAI's published sample clip, retrieved 2026-09-19. · Third-party material. Rights remain with the publisher; shown with source attribution and no claim of ownership.

A crested Spartan helmet on a battlefield at sunset with smoke, scattered figures and a long spear line receding behind it

The poster frame of the trailer that the announcement attributes to this model, made by David Thompson and published by xAI with the launch. This is a still, and it is the only image from that trailer this page can show: the trailer's own file is among the ones that answer 403.

xAI news, Grok Imagine Video 1.5 announcement, poster frame retrieved 2026-09-19. · Third-party material. Rights remain with the publisher; shown with source attribution and no claim of ownership.

A dark plain at night with a single standing figure silhouetted under a bright band of Milky Way stars

The still that xAI's image-to-video example feeds to this model, published by the publisher itself as the input for the prompt "Generate a slow and serene time-lapse" at a twelve-second duration. It is an input frame, so it shows what the model is asked to animate, not what it produced.

xAI documentation, image-to-video page, retrieved 2026-09-19. · Third-party material. Rights remain with the publisher; shown with source attribution and no claim of ownership.

Questions

Is this the official page for Grok Imagine Video 1.5?

No. It is a page on this site about the model. The publisher's own pages are linked as sources, and this page has not been reviewed or endorsed by xAI.

Which modes can I actually run here?

Text to video and image to video. The service also declares reference to video for this product, but that mode takes images rather than clips, so the workspace does not list the family in its reference-to-video menu.

Why does the duration list go past fifteen seconds when xAI documents a fifteen-second limit?

Because two different sources publish two different lists. xAI's API documentation states 1 to 15 seconds; this site's service publishes twenty-three options from 6 up to 30 seconds. The workspace follows the service, and this page shows both rather than picking one.

Can I get 1080p output here?

Not through the settings this page describes. The publisher documents and prices a 1080p tier for this model, and the service this site sells through publishes 480p and 720p only, with 480p as the default.

Do the videos come with sound?

The publisher says generated videos carry an audio track by default and that it can be switched off upstream. This workspace publishes no audio field for this product, so there is no toggle here either way.

How many reference images can I attach?

The service publishes a limit of one to seven for the two image-driven modes, and the publisher's documentation also states a maximum of seven reference images per request.

How is the charge worked out?

The server quotes it in points for the mode and settings you chose before the task starts. The catalogue publishes a floor and a tier table for this product, and there is no flat price to quote on this page.

Will text-to-video hand me the first frame as an image?

No. The publisher states that text-to-video generates a first frame and then animates it, and that the intermediate still is not returned. You send one request and get back a video.

Why does the gallery here have only three items?

Because the rest of the publisher's footage could not be retrieved. Every clip file on the announcement, preview and reference pages answers with an access-denied page, from a browser session as well as from a command line, so the page shows the material it could verify and records the rest as a gap.

Explore next