This review is based on documented features, verified pricing, public reporting, and community sentiment — not hands-on testing. See how we research →
Grok Imagine Video 1.5 arrived at a convenient moment. OpenAI shut down the Sora consumer app in late April 2026, the Sora 2 API is scheduled to sunset in September, and into that gap stepped xAI with a model that is fast, cheap, ships with synchronized audio, and — by one widely cited third-party measure — currently sits at the top of the image-to-video leaderboard. On the technical scorecard alone, it reads like an easy recommendation.
It isn't — and the sticking point isn't the 720p ceiling. The harder problem is a documented trust-and-safety failure that the leaderboard coverage tends to skip past. This review treats both halves honestly: what the model does well, what it can't do, and the moderation record that should weigh on whether you or your brand put it into a workflow at all.
Grok Imagine Video 1.5 is xAI's image-to-video model. You give it a still image plus a short text prompt describing the motion you want, and it returns a clip of up to 15 seconds at 480p or 720p — with native, synchronized audio generated in the same pass rather than bolted on afterward. Clip lengths run 5, 10, or 15 seconds, and extensions of roughly 6–10 seconds per pass are possible, though the risk of visual drift rises the longer you push a single generation.
One distinction matters up front, because the marketing around video models blurs it constantly: this is image-to-video, not text-to-video. xAI's own model documentation states that the 1.5 preview does not support text-to-video generation. You bring the starting frame; the model animates it. If your workflow assumes a pure "type a sentence, get a video" pipeline, this is not that tool.
Under the hood it runs on xAI's Aurora engine, an autoregressive system that generates frames sequentially. The practical payoff of that approach is steadier camera movement and subject stability across a clip, which is where a lot of cheaper video models fall apart. The model moved from preview to general availability on June 16, 2026, with a "wide release" announcement following on June 17. It's reachable through grok.com/imagine, on X, in the iOS and Android apps, and via the xAI API — though note the API model name still carries a -preview alias (for example, grok-imagine-video-1.5-2026-05-30), a small but telling sign of how recently this shipped.
The headline number people cite is its leaderboard position. In blind, preference-based testing on the Image-to-Video Arena, Grok Imagine Video 1.5 currently ranks first, with an Elo around 1,330 — roughly 52 points above its own 1.0 release — ahead of Sora 2, Veo 3.1, Seedance 2.0, and Kling in that particular comparison. We treat that as a third-party signal, not as proof and not as an input to our score. Arena rankings measure which clip a general audience prefers on everyday prompts; they don't measure the things professional production actually needs, like 1080p delivery, precise frame control, or industry-standard formats.
What the ranking does corroborate is that the raw output holds up well for short, casual clips. Two things stand out as real differentiators. First, the native synchronized audio: generating sound in the same pass removes a separate scoring-and-syncing step that competing pipelines usually require. Second, speed — xAI reports a 6-second 720p clip generating in around 25 seconds in fast mode, which is quick enough to make iteration practical rather than painful. The Aurora engine's frame-by-frame generation also tends to keep the subject and camera coherent across a clip, which is the most common failure point in cheap video generation.
Surrounding the model is a reasonable amount of workspace tooling — Projects, multi-agent runs, library search, and an Agent Mode — that makes it more than a single-shot generator and more practical for anyone producing clips at volume.
Price is the clearest argument in Grok Imagine Video 1.5's favor, and it's worth being specific about why. The free web tier gives roughly five generation credits per day with no X Premium subscription required — enough to evaluate the model honestly before paying anything. SuperGrok at $30/month raises those limits for regular users. For developers, the xAI API charges per second of output: $0.08 per second at 480p and $0.14 per second at 720p, which works out to about $4.20 per minute at the higher resolution, plus a $0.01 image-input charge. Audio is included at no extra cost. The model is also available through the third-party platform ImagineArt on a credit-based Pro plan.
| Plan | Cost | What you get |
|---|---|---|
| Free (web) | $0 | ~5 generation credits/day, no X Premium required |
| SuperGrok | $30/mo | Higher generation limits for regular use |
| xAI API — 480p | $0.08/sec | Pay-per-second, audio included, +$0.01 image input |
| xAI API — 720p | $0.14/sec | ~$4.20/min, audio included, +$0.01 image input |
| ImagineArt (3rd-party) | Credit-based | Access via ImagineArt's Pro plan |
The comparison is what gives those numbers weight. Sora 2 Pro runs around $30 per minute and Google's Veo 3.1 sits in the rough range of $9–24 per minute, so at roughly $4.20 per minute for 720p, Grok undercuts Sora 2 Pro by about 86%. With the Sora consumer app already discontinued and the Sora 2 API set to sunset in late September, the timing turns a price advantage into a straightforward availability advantage for anyone who needs cheap, sound-on short clips right now. Pricing and limits move quickly on new models — verify current rates at docs.x.ai and grok.com before you build a budget around them.
The technical ceilings are easy to state. Output tops out at 720p, where several competitors deliver 1080p; xAI has reportedly floated a higher-resolution "Pro Mode," but with no date attached, that's a roadmap item, not a feature you can plan on. The model is image-to-video only — there is no reliable text-to-video path — so it slots into a workflow rather than replacing one. Clips are short, the API still wears a "preview" label, and as a newly generally available product it has a limited public track record to judge reliability against.
The subtler limitation is the gap between what the leaderboard measures and what professional work demands. Topping a preference arena on general prompts is not the same as hitting 1080p deliverables, holding precise frame-level control, or fitting into established post-production formats. For social clips and rapid concepting, the constraints are minor. For client or broadcast work, they're the whole conversation.
This is not a contained incident. Grok Imagine launched with a deliberately permissive "Spicy Mode" positioned as largely unfiltered, and the misuse emerged in that context — a system built to be permissive, not one that simply slipped. What surfaced in reporting in late December 2025 and early January 2026 has since been documented as an ongoing pattern: NBC News reported in April 2026 that non-consensual sexualized image generation continued as users worked around the platform's safeguards; a WIRED investigation in June 2026 found Grok still hosting non-consensual deepfakes; and a federal class action filed in California in March 2026 alleges manipulation targeting images of apparent minors. These accounts were reported by NBC News and WIRED, and are part of ongoing legal and regulatory scrutiny — among the most serious failures a generative media tool can have. We are attributing them and providing no detail about the content or how any of it was done.
For balance, xAI has taken documented steps. It restricted image generation to paid subscribers on January 9, announced anti-editing measures on January 14, says it reports suspected child sexual abuse material to the National Center for Missing & Exploited Children (NCMEC), and points to monitoring and prompt filtering. Those measures have been judged insufficient, however — Canada's privacy commissioner concluded that xAI failed to implement appropriate safeguards from the outset and was not convinced of their effectiveness, and the failures documented by NBC and WIRED postdate the fixes. The platform is under active regulatory scrutiny in multiple jurisdictions, including under the EU's Digital Services Act and by the UK's Ofcom, alongside the U.S. class action — proceedings that are ongoing and whose outcomes are not yet settled.
None of this is a comment on the technical quality of the video model. It is a trust assessment. A tool can produce strong output and still carry a moderation history that makes it unsuitable for a brand, an organization, or an individual who cannot accept that risk. For many readers, that record will be disqualifying on its own — and that judgment is entirely reasonable.
This record has also become relevant well beyond xAI's own products. SpaceX — which merged with xAI earlier in 2026 — has agreed to acquire Cursor's parent company, a deal announced in June 2026 and expected to close in Q3 pending regulatory approval, which puts the same organization's trust-and-safety record into a lot of developers' tooling decisions. We work through what that does and doesn't mean in our analysis of the SpaceX/Cursor acquisition.
Our score is the equal-weight mean of five dimensions. The output, ease, and value numbers are strong; the Trust & Maturity dimension is scored severely to reflect the documented moderation failure and the model's newly generally available status. The arena ranking is context, not a rubric input.
It's a fit for creators and developers who need cheap, fast, sound-on short clips generated from still images, and for teams reacting to Sora's exit who can work within a 720p ceiling — provided they have weighed the platform's moderation record and decided they can accept it. A typical job it handles well: animating a product still into a 6–10 second social ad with synced ambient audio, generated and iterated in a single sitting. At the free tier and at API rates this low, the cost of evaluating it is close to zero, which makes a scoped trial easy to justify.
Skip it if you need 1080p or higher, long-form clips, reliable text-to-video, or precise frame-level control — or if you simply cannot accept the platform's documented moderation history for your brand or use case. For higher-fidelity creative production, Runway remains the stronger choice; for avatar and talking-head business video, Synthesia and HeyGen are built for that job.
Free web tier available — about five generations a day, no X Premium required.
Visit Grok Imagine →On raw output, speed, audio, and price, Grok Imagine Video 1.5 is a strong image-to-video model, and its arrival as Sora exited makes the value case obvious. But the score and the recommendation are held back by two things the leaderboard coverage glosses over: a hard 720p ceiling with no text-to-video, and — far more seriously — the documented content-moderation failure detailed above. For many teams and brands, that safety record is disqualifying regardless of price.
The honest summary: technically impressive and inexpensive, but adopt only after weighing a serious, well-documented trust problem. The 7.5 isn't a 9-tier tool with an asterisk — it's a capable, cheap model whose composite is deliberately pulled down by a trust dimension that outweighs its arena ranking.
Weighing it against the alternatives? See our side-by-side Grok Imagine Video vs Runway vs Sora comparison for use-case routing across the three.