Guess I'll post what I learned about Grok Imagine while producing this. 

Grok is extremely accessible, you don't need a huge prompt to achieve what you want -- unlike some other models. 

The most effective way to achieve a visually appealing output is to let your starting frame do most of the talking.

From what I can tell, it reads prompts fairly linearly.

It's best to direct your characters by a core defining trait, or an @imageN inline chip (if you're using more than one reference), instead of saying every tiny detail about them. 

The moment you declare an object in a shot, the model will attempt to produce it.

For example, instead of stating the environment and all props up front, you should describe them as they appear in your scene:

" [0-4] Slow zoom in on the back of her head,

The duelist in the torn haori holds a curved blade low at her right side, centre frame, back to the storm-dark city, rain sheeting off the clay tiles at her boots, lightning cracks the sky

[4-6] Snap cut, extreme close up to the tip of the blade,

 she flicks blood from it upwards in one sharp arc, the water breaking into spray off the steel "

Notice how in the output attached below, it produces EVERYTHING I prompted, with near-perfect adherence to the timestamp provided.

Alternatively, you can use overload your prompt with a lot of detail to achieve chaotic, faster paced scenes, albeit with slightly less environmental texture. These types of prompts tend to be behave more like slot machines, however.

This is only a sample of what I learned while creating my submission, but this post will end up being 20 paragraphs long if I keep going. 

TL;DR

Grok Imagine is fun, has excellent prompt adherance, and is extremely easy to jump into and use.

@grok @imagine @XCreators thx pls watch my submission quoted below ty ?