What happened
On Oct 27 (late afternoon, US Central time), the official Grok Imagine account on X published a prompting guide as an X article called "Inside Episode 1: The Oath". It walks through how the team and Wonder Studios made the first episode of their Odyssey series, which they posted on Oct 23 and built with a crew of 3-5 people.
The stated aim is to get honest feedback from filmmakers to improve Grok Imagine's image and video models, so the guide reads as a working method rather than a feature launch.
The method, in order
Plan before generating. Write a shot list with shot size, lens, camera movement, action and exact dialogue for every shot, and keep folders named by scene. Put every frame, prompt and take on one Grok Imagine canvas so you can compare takes and check faces, wardrobe and lighting across a scene.
Lock the look. Make a simple character sheet for each character (a close-up, a full front and a full back on plain grey) and a new sheet if the costume or age changes. Name the camera, lens and film stock in every prompt. Get each still keyframe right before animating it, because a weak still makes a weak clip.
Write video prompts in seven blocks: camera, action, environment, physics, audio, SFX and constraints. The guide's advice when a prompt fails is to make it shorter, not longer. It also points to middle keyframes: you can add a frame between the start and end frames and choose when it lands, which the team used to time an entrance.
What it doesn't tell you
The guide is a post on X; we found no matching page on xAI's news site or docs as of Oct 28. It doesn't say which Grok plans include the canvas or middle keyframes, what generation limits apply, or how many takes the episode needed. It admits faces still drift sometimes, and the finishing work (about 18 hours of editing, sound and color, with music added outside Grok Imagine) is a real cost to plan for.
How to use it
If you're starting a short series, our Grok Imagine series playbook follows the same order: script, character sheets, keyframes, then shots, with interview-style prompts at each step. The seven-block structure also works for single ad shots in the AI video ad playbook.