Quick answer
Grok Imagine is xAI's image and video generator, built into Grok and X and sold through APIs. The current video model, Grok Imagine Video 1.5, makes 1 to 15 second clips with sound at up to 1080p, and it is the cheapest decent video model around: about $0.02 per second at 720p through resellers like KIE. It is fast and good at bringing a single photo to life, but it mostly knows one camera move (a slow push-in), sits mid-table on quality leaderboards, and has a history of sudden moderation changes. Use it for cheap drafts and simple one-shot scenes; use Seedance or Kling when the camera move or the acting is the point.
Grok Imagine is the cheapest video model that's actually usable in late 2026. That's both its pitch and its limit. Here's what it does well, where it breaks, and how to prompt it, from running it in production at Starrd.
Facts and prices are as of November 19, 2026.
Grok Imagine at a glance
| Grok Imagine Video 1.5 | |
|---|---|
| Maker | xAI |
| Released | Preview May 30, GA June 16, 2026 |
| Length | 1 to 15 s (default 8), extendable |
| Resolution | 480p, 720p, 1080p |
| Sound | Yes, with lip-sync |
| Inputs | Text, up to 7 images |
| Price (KIE) | $0.012/s 480p, $0.0225/s 720p, $0.04/s 1080p |
| Price (xAI API) | ~$0.08 / $0.14 / $0.25 per s, cheaper Lite tier |
| Leaderboard | ~#17 text-to-video (Artificial Analysis) |
Sources: xAI docs, KIE, Artificial Analysis.
There's also Grok Imagine Image 2.0 (August 2026) for stills, about $0.02 an image through resellers and #10 on the text-to-image board. It's covered in the best AI image generators.
What it's good at
Price. Grok bills a flat rate per second, so a 10-second 720p clip is about 23 cents on KIE. Seedance 2.5 costs about $3.15 for the same clip. If you're testing twenty ideas, that difference decides your budget.
Speed. Our renders come back in about 25 seconds. Seedance takes minutes.
Bringing one photo to life. With a single image, Grok treats it as the literal first frame and animates from there. That's exactly right when the photo is the scene, like your own living room. It's why Starrd's Christmas templates use it: you photograph your tree and Santa shows up in your actual room.
Catch Santa at Your Tree: your own room photo, animated by Grok Imagine 1.5. 100 credits.
Make thisMore Grok Imagine templates: the Grinch, the elf, Santa at the cookies
Where it falls apart
It has one camera move. We measured it on real-estate shots: asked for a push in, a slide, a crane up and a pull out, Grok got 1 of 4. Every miss was a slow push-in. Plain-English direction doesn't fix it.
Multi-shot scenes. We re-rendered our car templates on Grok against Seedance. Grok lost clearly. It ignores "this photo is a reference, not the opening frame" and opens on the photo anyway.
More than one image changes the rules. Give it two or more images and it switches to reference mode, with no first-frame guarantee.
Hard physical action. Santa coming down a chimney in a real room photo took four tries and never worked. He fell from the ceiling or stood on the mantel. Seedance 2.0 got it on the first try.
Moderation whiplash. After the 2025 deepfake scandals, users report sudden clampdowns and paid videos deleted without warning (PiunikaWeb, March 2026). Expect a prompt that worked last month to be refused.
Random failures. In batches we've seen every job fail with "server exception" and then all succeed on resubmit. Build in retries.
How to prompt Grok Imagine
- Describe the motion, not the photo. The image already shows the room and the person. Spend your words on what happens: who moves, what they do, what we hear.
- Ask for one camera move, ideally a push-in or a locked camera. Don't plan a crane shot.
- One beat per clip. Grok holds one action well. For a sequence, make several short clips and cut them together. Billing is per second, so twelve 2-second clips cost the same as four 6-second ones.
- Write the sound. Name the sounds you want ("a spoon clinks, the fridge hums") or it guesses.
- Crop your photo to the frame you want first. The output takes the photo's exact shape, whatever aspect ratio you ask for. A 3:2 photo gives you a 3:2 clip.
- Stay under about 4,000 characters. That's the cap on the 1.5 model.
@photo1 stands in their kitchen holding a coffee mug. Slow push-in. They look up, raise an eyebrow at the camera and take a sip. Morning light through the window, the hum of a fridge, a spoon clinks.
Try this prompt in Starrd Custom on Grok Imagine: 30 credits for 10 seconds. For more camera ideas, the camera movements library has copy-ready prompts.
Grok Imagine vs the alternatives
| You want | Use |
|---|---|
| Ten cheap drafts of an idea | Grok Imagine |
| A specific camera move | MiniMax H3 / Hailuo |
| A 30-second scene that holds together | Seedance 2.5 or Kling 4.0 |
| You in a trend video, no prompt | Starrd templates |
| To fix a clip by asking | Gemini Omni Flash 1.2 |
The full field, with prices, is in the best AI video generators of 2026.
Catch Santa at Your Tree
Photograph your tree. Santa shows up in your real room. 100 credits.