SpaceXAI: Grok Imagine Video

SpaceXAI: Grok Imagine Video is a video generation model from xAI. It sits in the video category and is built for turning text and image inputs into short video clips. If you are evaluating models for a pipeline that needs generated motion rather than text or still images, this is the kind of model to consider, and it comes from the same provider behind the Grok family.

On the input side, the model accepts text and images. On the output side, it produces video. The provider describes it as fast and as text-, image-, and reference-conditioned, which means you can drive generation from a written prompt, from an image, or from a combination of the two. Output clips run from 1 to 15 seconds at 24 frames per second, at either 480p or 720p. Seven aspect ratios are available: 16:9, 9:16, 1:1, 4:3, 3:4, 3:2 and 2:3. A first_frame image can be supplied, which lets you anchor the opening of the clip to a specific still. No context length is listed for this model, so treat the prompt as a generation instruction rather than a long conversational context.

Those capabilities point to a fairly specific set of uses. Short vertical clips in 9:16 suit social formats, while 16:9 and the wider ratios fit landscape playback. Because durations top out at 15 seconds, the model is a natural fit for single-shot sequences, animated stills, product or scene previews, and quick concept visualisation rather than long-form narrative video. Supplying a first_frame image is useful when you need continuity with existing artwork, a storyboard panel, or a photograph, and the range of aspect ratios means the same prompt can be adapted to different placements without regenerating from scratch.

You call the model through api14's OpenAI-compatible API using the model id "x-ai/grok-imagine-video". The supported request parameters are logprobs, max_tokens, response_format, seed, temperature, top_logprobs and top_p. The seed parameter is worth noting for video work, since it gives you a way to make runs reproducible when you are iterating on a prompt. Temperature and top_p control sampling, and the logprob-related parameters let you inspect token-level likelihoods where the endpoint returns them. Because the interface is OpenAI-compatible, existing clients and SDKs generally work with only the base URL and model id changed.

There are limits to keep in mind. Clips are capped at 15 seconds and resolutions at 720p, so work that needs longer runtime or higher fidelity will need a different model or an assembly step that stitches multiple generations together. Audio generation is not specified in the available details, so do not assume the output includes sound. If your task is text, reasoning, or still image work rather than motion, a model in those categories will fit better than this one.