Does Grok AI generate videos?
Yes. Grok Imagine makes video with synchronized sound — launched July 28, 2025 with 6-second clips, now up to 15 seconds via the API at 480p, 720p, or 1080p. Video 1.5 (May 31, 2026) renders a 6-second 720p clip in about 25 seconds. It is no longer free: xAI removed the last free image access on March 19, 2026.
Why — the first-principles explanation
A video model is not a chatbot with a camera. It's a denoiser. Training shows it millions of clips that have been progressively destroyed with visual static, and the model learns to undo that destruction one step at a time. Point it at pure noise, whisper a description, and it hallucinates its way back to a clip that matches the words. That's why AI video has its signature failure modes — fingers that merge, text that turns to gibberish, objects that quietly morph. It isn't recording anything. It's guessing what a frame should look like, over and over.
That guessing is brutally expensive, which explains every constraint on the product. Each second is dozens of frames, and every frame must stay consistent with its neighbors or the clip visibly boils. Cost scales roughly with pixels times seconds — so a 15-second 1080p clip is many times the compute of a 6-second 480p one. This is exactly why the price is quoted per second of output rather than per clip, and why free access disappeared. You can see the arithmetic in xAI's own API: `grok-imagine-video-1.5` bills $0.080 per second, the older `grok-imagine-video` $0.050 per second, while a still image costs $0.02. Video is roughly four seconds' worth of images per second of footage.
The recent jump is in speed and sound. Grok Imagine's early clips were six silent-ish seconds. Grok Imagine 1.0 (February 2, 2026) pushed to 10 seconds at 720p with much better audio. Video 1.5 (May 31, 2026) took the top of the Image-to-Video Arena leaderboard with a 52-Elo jump over 1.0, and cut a 6-second 720p render from over 40 seconds to about 25. Crucially, sound effects, ambience, and dialogue are generated in the same pass as the picture — not dubbed on afterward — which is why the audio lands on the action instead of sliding off it.
One honest caveat that most pages skip: the loose filters that define Grok apply here too. In December 2025 users drove Grok Imagine's tools into producing nonconsensual sexualized images of real women and minors, triggering international backlash and investigations. The capability is real; so is that history.
An example that makes it click
Imagine a sculptor handed a block of marble that is pure random scratches, and told: "there's a golden retriever catching a frisbee in there — find it." She can't check a photo. She just keeps chipping wherever the scratches look least dog-like, until a dog appears. That's the model.
Now ask for 6 seconds of dog. At 24 frames a second, that's 144 blocks of marble, and every dog must be the same dog, mid-jump, in the right position. That's why you pay by the second, why 1080p costs more than 480p, and why the free tier died: each clip is 144 sculptures that have to agree with each other.
How to do it
- Open grok.com/imagine, or the Grok app on iOS or Android, or the Imagine tab inside X.
- Pick your starting point: type a text prompt, or upload a still image — image-to-video is what Grok's 1.5 model is strongest at, and 1080p is limited to image-to-video on version 1.5.
- Write the prompt with camera and motion in it, not just nouns. 'Slow dolly-in, handheld, rain on the window' gives the denoiser something to aim at; 'a cool city' does not.
- Choose duration and aspect ratio. The API allows 1–15 seconds and supports 1:1, 16:9, 9:16, 4:3, 3:4, 3:2 and 2:3. Video editing is capped at 8.7 seconds.
- Generate, then re-roll. Output is stochastic — the fastest path to a usable clip is three cheap attempts, not one perfect prompt.
- For volume or automation, call the Imagine API directly with grok-imagine-video-1.5 and budget $0.080 per second of output (as of 2026-07).
Key facts
- Grok Imagine launched July 28, 2025, generating six-second clips from text prompts.
- Grok Imagine 1.0 (February 2, 2026) raised output to 10-second videos at 720p with substantially improved audio.
- Grok Imagine Video 1.5 launched May 31, 2026, taking first place on the Image-to-Video Arena leaderboard with a 52 Elo point jump over version 1.0; a 6-second 720p clip renders in roughly 25 seconds, down from 40+.
- The xAI API allows video durations of 1–15 seconds at 480p (default), 720p, or 1080p, with 1080p limited to image-to-video on version 1.5; video editing is capped at 8.7 seconds.
- API pricing as of 2026-07: grok-imagine-video-1.5 costs $0.080 per second, grok-imagine-video $0.050 per second, grok-imagine-image $0.02 per image, grok-imagine-image-quality $0.05 per image.
- xAI removed the last free Grok Imagine image access on March 19, 2026; video generation now requires a paid tier.
▶ The 60-second explainer (script)
Yes — Grok makes video, through a tool called Grok Imagine. It launched in July 2025 with six-second clips. Today the API will give you anywhere from one to fifteen seconds, at 480p, 720p, or 1080p, with sound generated in the same pass as the picture — which is why the audio actually lands on the action instead of sliding off it. The newest model, Video 1.5, shipped May 31, 2026 and took first place on the image-to-video leaderboard. It renders a six-second 720p clip in about twenty-five seconds. Here's the part worth understanding. A video model isn't filming anything. It's a denoiser — it starts with pure visual static and chips away at it until something matching your words appears. Do that twenty-four times per second, and every frame has to agree with its neighbors. That's brutally expensive, and it's why xAI charges by the second: eight cents per second for version 1.5, five cents for the older model. A still image? Two cents. And it's why the free tier is gone — xAI cut the last free image access on March 19, 2026. One thing other pages won't tell you: Grok's loose filters apply here too. In December 2025, users pushed these tools into making nonconsensual sexual images, and regulators noticed.
What authoritative sources say
People also ask
Can I generate Grok videos for free?
No. xAI removed the last free Grok Imagine image access on March 19, 2026, and video generation sits behind a paid tier. The API bills per second of output.
How long can a Grok video be?
The API accepts 1 to 15 seconds per generation, with video editing capped at 8.7 seconds. Consumer app limits are lower and vary by subscription tier.
Does Grok Imagine generate sound?
Yes — sound effects, ambience, and dialogue are generated in the same pass as the video, which is why they sync to the action rather than being dubbed on afterward.
Is text-to-video or image-to-video better?
Image-to-video. Grok's 1.5 model is explicitly an image-to-video model, and 1080p output is limited to that mode. Starting from a still gives the model a fixed subject to keep consistent.
Why do my clips look melty?
Because the model guesses each frame rather than recording it, and small guessing errors compound across frames. Shorter clips, simpler motion, and starting from an image all reduce it.