Seed consistency and variations of Firefly video output

I would like Firefly video to allow subsequent output of content to be consistent in look and feel with, so new content with clearly specified (by the user) variations. This is possible to a degree but has limitations as I'll cover.

Examples of what I've tried to do with limited success.
* Create a powerful looking character by the camera angle and the expression in the character. Then have the same character looking vulnerable. 
* Create character in video doing one thing for a few seconds, then the same character doing something different for a few seconds.

In the current beta version, changing the prompt slightly can give significantly different looking results.

You can use a reference image but it is restrictive. As mentioned in my post Firefly Video 'Shot size' and 'Camera angle' with reference image, if you use a reference image, you can't change the camera angle by the settings or in the prompt.

Firefly Video has a seed number setting. My understanding of a seed is that if you use the same seed, prompt and settings, you would get the same image every time. This has not been the case with me. I'm not sure if this is a technical issue or me not understanding the nature of the Firefly video seeed setting. See the images below with the exact same prompt, same seed, a different looking results.

Elephant example (short prompt)

Example 2 (long prompt)

My experience, which I believe is a challenge with any Gen AI image/video tool, is that the more complex and longer the prompt, the less consistent, as illustrated above.

I have had some success with the same prompt but different camera angles, as in the images below. The main subject looks very similar. There are slight differences in the background.

Firefly is in need of much more settings to allow you to decide what to lock in, what you'd like to be changed, and when to go new vs variation.