Skip to content

Video

Grok Video 1

Video

Available on Ofrai — chat, images, and video in one place with shared credits.

46 creditsModel details

Video studio

Preview

No clip yet

Your latest clip preview will show here after generation.

Open studio

Example prompts

Choose an example to load its prompt and recommended settings into the Grok Video 1 creator above.

How it works

Create with Grok Video 1 in three steps

  1. 01

    Choose your inputs

    Keep Grok Video 1 selected, then add a prompt and any supported reference assets.

  2. 02

    Set the output

    Choose the aspect ratio, resolution, and available Video options before generating.

  3. 03

    Generate and refine

    Review the result, adjust the prompt or settings, and generate another version from the same place.

Grok Video 1 is an endpoint whose entire request surface is four optional fields: prompt, duration, resolution and aspect ratio. That sounds like a stripped-down model, and in one sense it is — there are no numbered reference lists, no placeholders, no tail frame, and no audio parameter to negotiate.

But the optionality cuts the other way as well, because the documentation lists no default value for any of those four fields and recommends writing them in explicitly, which means nothing is filled in on your behalf and nothing is guaranteed if you leave it out.

Three generation paths share the surface: text-to-video, first-frame image-to-video, and reference-image mode, and they do not have identical ceilings — the path you choose decides which resolutions are open to you, and that is the single most useful thing to understand about this model before you spend anything on it.

Three paths, one flat set of parameters

There is no routing field here, which means the path is inferred from what you send rather than declared. Send text and you get text-to-video. Send an image alongside a prompt and you are in first-frame or reference-image territory, and the documentation treats reference-image mode as its own distinct path with its own resolution ceiling. Three practical consequences follow.

First, an empty prompt is not a legal way to express "just animate this" on every path: the documentation marks prompt as optional at the schema level while stating that text-to-video and reference-image generation require it in practice. Second, because nothing is validated by a mode field, a mistake shows up as a surprising result rather than as a rejected request — which makes it worth sending all four fields every time instead of discovering the inference rule by experiment.

Third, with no separate parameters for start and end frames, a shot that must land on a specific closing image has no way to express that requirement here, and no amount of prompt writing will substitute for a field the schema does not expose.

Duration starts at one second, which is the real story

The documented duration range is a full second to fifteen seconds, as whole seconds, with no default listed and an explicit recommendation to set it yourself. Almost every video endpoint in this catalogue starts its floor at three seconds because a shorter render is rarely worth paying for. This one does not, and that single difference is more consequential than the fifteen-second ceiling.

At the short end you can produce motion fragments — a one-second product flick, a two-second loop-friendly beat, a transition tile that a compositor can cut into a longer sequence — which is a category of work most models cannot do at all because they refuse the request. At the long end, fifteen seconds is a continuous take, and past roughly ten seconds a single take starts to drift on identity and background.

The sensible planning rule that falls out of the documented range: use the short end for inserts and fragments that your edit will assemble, and use the long end only when the shot is genuinely one unbroken take.

Resolution is capped by mode, and the top tier is conditional

Three values are available — 480p, 720p and 1080p — but they are not equally available in every path, which is the detail most integrations get wrong. Reference-image generation is documented as working at 480p or 720p. Text-to-video and first-frame image-to-video may reach 1080p, and even there the documentation adds a condition: 1080p is available when that run's capacity allows it. That word "when" is doing real work.

Top-tier availability is not a constant, so a 1080p request that worked on one run is not a guarantee about the next one, and the right operational response is to treat 1080p as best-effort rather than as a resolved promise.

The reference page's own examples all pin 720p explicitly, which is a reasonable default to imitate for anything that has to reproduce: it is inside every documented ceiling, it does not depend on run-to-run availability, and it is the resolution your iteration renders should be using anyway.

Aspect ratio is five values, filtered by availability

The list is 16:9, 1:1, 2:3, 3:2 and 9:16 — a fuller set than most endpoints offer, since it includes both portrait and landscape three-by-two framings. The caveat sits in the same sentence: which of them are actually reachable depends on what is available on a given run. The practical reading is that the enumeration describes the shape of the parameter, while the subset you can rely on is negotiated at request time. Two habits follow.

First, generate at the ratio the composition was designed for rather than cropping afterwards — the vertical three-by-four and two-by-three framings exist precisely so that a phone-first composition is not a centre-cut of a widescreen render. Second, if a ratio you asked for comes back letterboxed or cropped rather than rejected, that is availability narrowing the set, not a bug in your prompt, and the fix is to request one of the more universal values rather than to keep retrying the exotic one.

Why you should send a prompt even when the schema says optional

This is the one place where the documentation is genuinely ambiguous, and it is worth reading closely rather than skimming. At the schema level, prompt is optional with no default. In practice, text-to-video and reference-image generation require it. For first-frame requests the documentation notes that some older single-first-frame requests were allowed to omit it, while other first-frame requests may still demand it.

Three states exist in one field description, and they resolve to a single piece of advice: if you are building a new integration, always send a non-empty prompt, which is exactly what the documentation recommends. The cost of following that advice is negligible, and the cost of ignoring it is a request that fails on a path you thought was prompt-free. What the prompt should contain differs by path. With text-to-video it carries everything, since nothing else is supplied.

With an image input, the frame is already fixed, so the highest-value sentences are about movement, pacing and camera — describing what is already visible in the frame competes with the image for the model's attention and is a common cause of a first frame that drifts a few seconds in.

Planning a sequence on an endpoint with no documented defaults

Because every field is optional and nothing has a published default, the reproducible request is the explicit one: prompt written, duration written, resolution written, aspect ratio written. Build a small template that always fills all four, and the output stops depending on what gets filled in for you by default. That same property shapes how you should work.

There is nothing to keep stable between runs — no reference index, no tail frame, no declared mode — so two calls with identical parameters are the only true baseline you have, and a second attempt is a re-roll rather than a controlled experiment unless you change one parameter at a time and record which. It also means there is no mechanism on this endpoint holding a face, a product or a costume consistent across two separate renders; multi-shot continuity has to come from your edit.

Render the shots as separate clips, assemble them downstream, and treat each one as if it were the only shot in the sequence.

FAQ

Which resolutions can I request on this model?
Three values are published: 480p, 720p and 1080p. Reference-image generation is documented at 480p or 720p, while text-to-video and first-frame image-to-video can reach 1080p when availability allows it. Pin 720p if you need a result that reproduces across runs, since it sits inside every documented ceiling and does not depend on run-to-run availability.
Does duration have a default value?
The documentation lists no default for it. The range is whole seconds from 1 to 15, and the documentation recommends setting it explicitly rather than relying on an implied value. The 1-second floor is the notable part: it allows motion fragments and loop-friendly inserts that most endpoints, which start at 3 seconds, will simply refuse.
Do I have to send a prompt?
It is marked optional in the schema, but required for text-to-video and reference-image generation, and the documentation notes that while some older single-first-frame requests could omit it, other first-frame requests may still require it. For a new integration, always send a non-empty prompt — that is the documentation's own recommendation.
How many aspect ratios are supported?
Five are enumerated: 16:9, 1:1, 2:3, 3:2 and 9:16. Which of them are reachable depends on what is available on a given run, so treat the list as the shape of the parameter and availability as the limit — if an unusual ratio returns cropped or letterboxed output, request a more universal value rather than retrying.

FAQ

Can I use Grok Video 1 directly on this page?
Yes. The creator above is the working Ofrai studio with Grok Video 1 selected by default.
Where are my finished results saved?
Signed-in generations are kept in your Ofrai library so you can download or revisit them.
Which prompts work best with Grok Video 1?
Describe the subject, setting, visual style, lighting, and camera or motion clearly. The examples above provide practical starting points.
Can I edit the example prompts before generating?
Yes. Load any example, then change its wording, aspect ratio, references, or other supported settings in the creator.

Compare other Video models that can handle similar creative work to Grok Video 1.

Commercial use depends on the selected model and model terms.