The short version, checked against xAI’s own pages on 22 September 2026: image generation appears on the free plan, video generation does not, and xAI publishes no number for what “free” actually buys you. Most articles answering this question quote a confident daily quota. None of those numbers come from xAI.
Here is what the primary sources actually say, including one place where xAI’s pricing page contradicts itself.
What free gets you
The plan comparison grid on x.ai/pricing has a row for image generation and a row for video generation. On the Free column, image generation carries a checkmark and video generation carries a dash. So on that table, free means stills only.
The same screen then disagrees with itself. The Free plan card lists real-time web and X search, Voice mode, SOC 2 compliance and Connectors. Image generation is not among them. The SuperGrok card at $30 a month lists “Image and video generation” as one of the things you are paying for.
Both statements are on the same page. Treat any claim about free Grok Imagine, including this one, as perishable. Free access has moved at least twice during 2026 and the page has not fully caught up with itself.
| Plan | Price | Image | Video |
|---|---|---|---|
| Free | $0 | Yes, per the comparison grid | No |
| SuperGrok | $30/mo | Yes | Yes |
| SuperGrok Plus | $100/mo | Yes | Yes, including 1080p |
Why nobody can tell you the daily limit
Because there isn’t one. In June 2026 xAI replaced per-product daily caps with a single shared weekly pool. From the Grok FAQ: instead of separate daily limits for Chat, Imagine, Voice or Build, you get one weekly allowance to spend across any Grok product.
That pool is shown as a percentage in Settings and Usage, broken down by product, and xAI deliberately publishes no absolute figures. Different products consume different amounts of compute, so a video costs far more of your pool than a chat message.
Two consequences worth knowing. When you hit the cap you can buy Extra Usage Credits starting at $5, web only, expiring after a year. And 720p video silently falls back to 480p once you pass the 720p cap for your tier, which looks like a quality bug and is actually a quota behaviour.
One detail cuts against the pricing grid. The FAQ says that at the cap you keep “free tier limits on Chat and Voice”. Imagine is not named. If you want certainty about your own account, the honest answer is to open Settings and Usage and read the percentage, because that is the only number specific to you.
The six-second limit is out of date
A lot of tutorials still say Grok Imagine caps video at six seconds. That was the original consumer launch spec in August 2025. It is wrong now.
The current documented maximum is 15 seconds, at 480p, 720p or 1080p. The path was 6 seconds at 480p at launch, then 10 seconds at 720p with the 1.0 release in early 2026, then 15 seconds at 1080p today. Audio is native rather than bolted on, and since July 2026 you can pass a voice reference alongside a character image so the same face and the same voice persist across scenes.
Other current specs: up to 10 images per request at 1024 or 2048 square, image editing from one to three source images, video editing on a source clip up to 8.7 seconds, and up to seven video references per generation.
There is an official API, and it is not expensive
Plenty of posts still say the API is “coming soon”. It shipped. Both endpoints are live at api.x.ai, the SDK is OpenAI-compatible, and video generation is asynchronous: you submit, receive a request ID, and poll.
| Model ID | Price |
|---|---|
grok-imagine-image |
$0.02 per image |
grok-imagine-image-2.0 |
$0.04 per image |
grok-imagine-image-quality |
$0.05 per image |
grok-imagine-video |
$0.050 per second |
grok-imagine-video-1.5 |
$0.080 per second |
Text input is free, image pricing is flat regardless of prompt length, and video bills per second of output. A 15-second clip on grok-imagine-video-1.5 is $1.20.
One hard date to diary: grok-imagine-image-quality retires on 2 November 2026. After that, requests to that slug get served by grok-imagine-image-2.0 with quality set to low. If you pin that model string anywhere, change it before November.
API usage falls under the enterprise terms rather than the consumer ones, and xAI states that generated media is not used for training.
Watermarks are not optional
Every generated image and video carries a Grok watermark. There is no setting to remove it. The acceptable use policy separately prohibits stripping, altering or circumventing provenance metadata or watermarks, and xAI notes that labelling AI-generated content is a legal requirement in some jurisdictions.
This matters because a whole cottage industry of watermark removers advertises against exactly this. Using one puts you outside the terms of the service that made the video, and outside disclosure law in some places. It is not a grey area worth the time it saves.
On ownership, the terms updated on 11 September 2026 say you retain your rights to output. Two clauses complicate the clean answer people want. You are required to attribute generated output to the service per the brand guidelines, and there is no IP indemnification running in your favour. Outputs are also explicitly not guaranteed unique, so another user may generate something very similar.
About those “free unlimited Grok Imagine” sites
Search this topic and you will find mirror sites offering unlimited free generations. The acceptable use policy prohibits scraping, harvesting or reselling any input or output, and prohibits operating sites that offer violative outputs for payment.
That is not the same as saying every third-party route is illegitimate. xAI quotes fal on its own Imagine API page, so licensed partners clearly exist. The distinction is licensing, and it is not visible from the front page of a site promising free unlimited anything. If a service resells model access, the thing worth asking is whether its access is licensed, because an unlicensed reseller can lose the model overnight and take your integration with it.
What I would actually do
If you want to try it, use the free tier for stills and accept that the quota is a percentage bar rather than a number. If you are building anything, use the official API: at two to four cents an image it is cheap enough that the aggregator discount rarely justifies adding a dependency between you and the provider.
And if you are generating media as part of a longer coding or automation task, note that the API is OpenAI-SDK compatible and asynchronous, which makes it straightforward to call from inside an agent loop rather than from a browser tab. That pattern gets interesting once more than one agent is doing it, because the polling, the credentials and the output all need somewhere to live. That is the layer Sharkly works on, around the tools doing the execution.
