Grok Imagine: the free tier came back, with a hard cap

What Grok Imagine is

Grok Imagine is not a standalone product. It is a tab inside Grok — the assistant — and it covers four jobs with one prompt box: generate an image, edit an image in plain language, animate a still into video, and edit that video. The intended workflow is the one the interface pushes you toward: get a picture you like first, then make it move.

There are four doors into it, and they are not equivalent:

• grok.com/imagine — the web surface, and the fastest place to test what the free tier currently allows.
• The Grok app on iOS and Android — the same thing on a phone.
• Inside X — reachable from the Grok icon in the navigation bar, with availability depending on your plan and region.
• The API — the models behind an API key, which is the only route where the price is published as a number rather than as an allowance.

Who makes it now — because the name changed twice

Every guide still says xAI, and the documentation no longer does. The sequence matters if you are reading anything dated after mid-2026, or if you are trying to find the right pricing page:

What does not change is the thing you type in code: the model identifiers still begin grok-, and the product is still called Grok Imagine. If you are following an older guide, its technology descriptions are probably still accurate — its company name is not.

Is Grok Imagine free? The honest answer is a timeline

This question has three different correct answers depending on the date you asked it, which is why so much of what you find is contradictory. Here is the sequence, with the current position last.

The cap is where the detail matters. In the October test, selecting 720p, a 10 or 15 second duration, or attempting a second video in the same day all brought up the subscription prompt. The allowance reset the next day, and the pattern held across four consecutive days — which is what makes it look like a slot rather than a one-off trial.

What it costs

There are two completely separate price lists, and mixing them up is the most common mistake. On the consumer side you buy a subscription with an allowance inside it; on the API side you pay per second and per image, and the rates are published.

The API rates, per second of video

Read down the 720p column and the shape of the market becomes clear: the lite model is roughly five times cheaper than the current one at the same resolution, and the older model sits in between. Ten seconds of 720p costs about $0.30 on lite, $0.70 on the older model and $1.40 on the current one. Since most projects need several attempts per usable shot, that multiple compounds fast — the honest first question is not "which model is best" but "do I need audio input and 1080p, or do I need a cheap first draft".

The subscription tiers

The vendor does not publish how many videos a plan contains. Third-party tracking puts the tiers roughly here, and the numbers move:

Which video model to use — and the capability 1.5 dropped

Three video models are live, and the newest one is not the right answer to every question. Two differences matter more than the version numbers:

• 1.5 dropped video input. Reference-to-video — supplying a clip for the model to work from — and video-to-video editing are listed only for the older grok-imagine-video model. If your workflow starts from existing footage rather than a still, the older, cheaper model is the one that supports it.
• 1.5-lite is the draft model. It is the only one priced for iteration, and the only one that cannot take an audio reference. For blocking out a shot before committing to a final render, it is the sensible default.

The practical split most people land on: draft on lite, finish on 1.5 when you need 1080p or supplied audio, and keep the older model for anything that starts from a video rather than a photo.

One more thing about the API. The preview model page for text-to-video states that mode is not supported in the preview — meaning image-to-video is the route that works today if you are generating programmatically, even though the consumer app offers text-to-video in its interface. Check the model page before you design a pipeline around a prompt box.

Limits, allowances and artefacts worth knowing

• The weekly pool. From June 2026 the vendor replaced separate daily limits for Chat, Imagine, Voice and Build with a single shared weekly usage pool. Video draws that pool down far faster than chat does, so heavy video use reduces what is left for everything else on the same plan.
• Silent downgrades. The documented behaviour is that 720p generations fall back to 480p automatically once you hit your plan's 720p allowance — without a warning at the moment it happens. If quality drops mid-session, that is why.
• Length. Up to 15 seconds per generation, fixed 24 fps, in portrait, landscape or square. Video editing keeps the source clip's duration and is capped at 8.7 seconds, so you cannot use editing to extend a clip.
• Reference images for editing. The image editor accepts up to three reference images, which is how character and product consistency is maintained across a set.
• Watermarks — unresolved. One reference work states all consumer output carries a Grok watermark that cannot be switched off; a pricing breakdown states free output is watermarked and a paid plan removes it. We could not settle this from primary sources. Check your own export.
• Region. The product is unavailable in mainland China, and availability inside X depends on your plan and region.
• New keyframe control. The web release notes for 3 October 2026 describe video mode treating a single attached image as the first frame, with a separate slot for a last frame. A community roundup dated 4 October reports intermediate keyframes had appeared in the interface without an announcement — pinning images at the start, the end and points in between, with the model filling the motion. That is a meaningful change for multi-shot work, and it arrived without a launch post, so treat it as observed rather than specified.

Where this page could be wrong

Three areas of this page rest on sources that disagree with each other, and it would be dishonest to present them as settled. We are listing them rather than picking a winner.

There is also a launch-date discrepancy worth naming, because you will meet it: one aggregator dates the video model's release to 29 January 2026 while the product history points to a consumer launch in August 2025, a developer API in January 2026 and version 1.0 in February 2026. Those are probably measuring different events — API availability versus model release — but no source reconciles them.

Getting a usable clip

The workflow that produces something publishable is shorter than most guides suggest, and the ordering is the whole trick.

1. Start from a sharp, front-lit still.

   Grok Imagine is designed image-first. A soft or busy source photo animates badly, and if you are putting a person in the clip, expect the face to drift a little across frames — a sharp front-lit still and a short clip hold a likeness far better than a long, busy one.

2. Describe one change per attempt.

   The most common cause of an off-target generation is asking for a complex scene plus several edits at once. Change one variable per retry and you learn what the model is actually responding to.

3. Write the motion, because that is the part you are not showing it.

   Slow zoom, pan left, character turns to speak. The still already carries the subject and the look; the prompt only has to carry the movement.

4. Draft at 480p, finish at the resolution you need.

   Cost scales with resolution and duration on every route. Block the shot out cheaply — on the API, the lite model is the tool built for exactly this — and only re-render the take you are keeping.

5. Check the audio if you turned it on.

   Synchronised speech is the headline feature and also where errors hide. If you cannot judge the lip-sync in the language being spoken, get someone who can to watch it before it ships.

6. Decide about the watermark before you commit to a plan.

   Because sources disagree about whether a paid plan removes it, check the export on your own account before you buy an upgrade on the assumption that it will.

Frequently asked questions

What is Grok Imagine?

The image and video generation feature inside Grok, the assistant built by the company formerly called xAI and now named SpaceXAI. It is not a separate product: it lives at grok.com/imagine, in the Grok iOS and Android apps, inside X, and as models behind the vendor's API. Two families sit underneath — Grok Imagine Image for stills and editing, Grok Imagine Video for clips — and both are reachable programmatically.

Is Grok Imagine free?

It has been free, paid, and free again. It launched effectively free in August 2025, was restricted to paying subscribers in January 2026, and moved out of the free tier entirely around 19 March 2026. In early October 2026 it returned: a four-day hands-on test found one free video per day at 480p and six seconds, with free aspect ratio and sound choice, plus free image generation. 720p, longer durations and a second same-day video trigger the subscription prompt. That figure comes from testing rather than a published spec, and other guides claim much larger allowances — confirm on your own account.

How much does Grok Imagine cost?

The consumer side is bundled into subscriptions with an unpublished allowance inside; third-party tracking puts the tiers from about $10 to $100 a month, with 1080p video only on the higher ones. The API side publishes real numbers: $0.08, $0.14 and $0.25 per second at 480p, 720p and 1080p on the current video model, against $0.02, $0.03 and $0.14 on the lite model, plus about $0.01 per input image.

How long can a Grok Imagine video be?

Up to 15 seconds in a single generation, at up to 1080p on the current model, at a fixed 24 fps. Community testing of chaining reports visible drift after two or three extensions, so longer sequences need care. Video editing keeps the original clip's duration and is capped at 8.7 seconds, so it will not extend a clip.

Does Grok Imagine make video with sound?

Yes, in the same pass as the picture rather than dubbed afterwards — dialogue with matching lip movement, ambience and effects. The current model also accepts an audio input, so you can supply a voice reference. The lite model does not accept audio input.

Does Grok Imagine have an API?

Yes, with published rates. Video: grok-imagine-video-1.5, -1.5-lite and the older grok-imagine-video. Images: grok-imagine-image, -quality and -2.0. Generation is asynchronous — submit, then poll. Note that the preview text-to-video model page states that mode is not supported in the preview, so image-to-video is the dependable programmatic path today.

What is the difference between grok-imagine-video and grok-imagine-video-1.5?

Newer is not simply better. 1.5 adds 1080p and audio input but costs twice the older model at 720p and nearly five times the lite model. It also dropped reference-to-video and video-to-video editing, which are listed only for the older model. Choose by what you need — 1080p and supplied audio on 1.5, or cheaper output and video-input editing on the older one.

Who makes Grok Imagine — is it still xAI?

No. SpaceX's acquisition of xAI closed on 2 February 2026 and the unit was rebranded SpaceXAI in July 2026; on 4 October 2026 a further rename to SpaceXSI was announced. The developer documentation already carries the SpaceXAI name while model identifiers still begin grok-. Older guides are right about the technology and out of date on the company name.

Is there a watermark on Grok Imagine output?

Sources disagree and we could not settle it. One reference work says all consumer output carries a Grok watermark that cannot be turned off; a pricing breakdown says free output is watermarked and a paid plan removes it. API output carries no consumer watermark. Check your own export before upgrading on the assumption that a plan removes it.

What is Spicy mode in Grok Imagine?

A content-rating setting that relaxes the output filters, behind a paid plan. It also has the most visible history in this product: the feature was used to mass-produce abusive images, and that fallout is why image generation was restricted to subscribers in January 2026. A sizeable slice of the search demand around Grok Imagine is looking for uncensored alternatives, which is why adult platform names appear in its related searches. We do not cover that use and we do not publish prompts for it.

Can I use Grok Imagine for commercial work?

Only within the terms of the surface you generate on — the app, X and the API each carry their own. Two practical points: free-tier output may be watermarked, which usually rules out client work, and post to X is governed by X's synthetic-media rules, which prohibit deceptive fabricated media of a real person. Generating inside Grok does not exempt the post from those rules, and enforcement lands on the account that posts it.

Also on this site

• Seedance 2.5 — the 30-second model from ByteDance Seed, and the one that most often gets compared with this.
• Kling 4.0 — the other competitor that undercuts on price. What Kuaishou has shipped versus announced.
• MiniMax H3 — ranked first on Arena's blind image-to-video board and downloadable, which neither of the above is.
• Free AI video tools — what "free" actually buys across every kind of generator, and what to check before you pay.
• Image to video AI — the four-step workflow and the prompt pattern that stops faces morphing.
• AI video tool status tracker — which tools still serve a working product, checked against their own domains.