# Generating video with Grok: a hands-on guide

By the GO AI Team · August 21, 2026

> **TL;DR:** Grok's video model turns a sentence — or a photo — into a clip of up to **15 seconds**. In [GO AI Chat](https://goaichat.app/chat) it lives in the **Lab** tab next to image generation. The prompt pattern that works: *one subject, one action, one camera move, one style*.

## What it is

Text-to-video and image-to-video from xAI. You give a description or a still; it returns a short clip with synthesised motion. It is a **shot generator** — one moment, a few seconds long — not an editor and not a scene-builder. The same capability appears in other surfaces (X, for one); in GO AI Chat the output goes to your camera roll rather than a feed.

## The four-part prompt

- **Subject** — specific. "A woman in a yellow raincoat," not "a person."
- **Action** — exactly one. "Walking toward the camera."
- **Camera** — the model moves it whether or not you ask, so ask. "Slow push in."
- **Style** — "shot on 35mm, shallow depth of field."

Assembled: *"A woman in a yellow raincoat walks toward the camera through neon-lit rain, slow push in, shot on 35mm with shallow depth of field."*

## Settings that matter

| Setting | Pick | Why |
|---|---|---|
| Duration | 5–8s usually; 15s only when needed | Longer clips drift. A tight 6s beats a wandering 15s. |
| Aspect ratio | 9:16 for Reels/TikTok/Shorts, 16:9 otherwise | Generate in the shape you'll publish in. |
| Quality | Low while iterating, high for the keeper | Prompt exploration is where runs get spent. |

Longer and higher-quality clips cost more tokens, so choose deliberately.

## Image-to-video

Starting from a photo removes the hardest part: the model no longer invents what things look like, only how they move. The rule that decides success — **the still must imply one plausible motion**. Loose hair implies wind; coffee implies steam; a parked car implies nothing.

## Genuine weaknesses

- **Text in frame** — signage and captions come back as gibberish. Add text later.
- **Hands doing fine work** — typing, instruments, small objects.
- **Character consistency across clips** — two generations won't match.
- **Counting** — "three dogs" often yields two or four.

## A workflow that wastes fewer runs

1. Write the four-part prompt; delete the second verb.
2. Generate once, low quality, short duration.
3. 80% right? Regenerate the same prompt — variance often fixes it.
4. Wrong in a specific way? Change exactly one part.
5. Re-run at the quality and length you want.

Step-by-step with the buttons: [iPhone video walkthrough](https://goaichat.app/blog/how-to-make-ai-videos-on-iphone). Grok outside video: [Grok 4 vs GPT-5](https://goaichat.app/blog/grok-4-vs-gpt-5).

---

We build [GO AI Chat](https://goaichat.app/chat) — GPT-5, Gemini, Grok 4, and DeepSeek in one iOS app, with Grok video generation in the Lab.

More: [GO AI Blog](https://goaichat.app/blog) · [support@goaichats.com](mailto:support@goaichats.com)