Method
There is no magic prompt
People who sell prompts and complicated chains promise results a prompt alone cannot deliver. Here is how I work.
I have been saying it since I started, and I say it as loudly as I can: there is no magic prompt.
Who sells it
Plenty of companies sell prompts, or pretend to believe that very complex prompts or very complicated strategies give better results. Nothing could be more wrong. In videos made with artificial intelligence, the prompt is almost never what makes the difference.
The cost of complicated strategies
Then there is something people rarely talk about. Very complex strategies, such as agentic systems with chained checks and interactions, are extremely expensive in tokens. For anyone on a standard, pro or even max plan they are practically unusable, because they consume a huge amount.
How I work
The prompt matters, of course. But honestly, I often use very simple prompts, especially when I animate an image and turn it into video. I do not use complex chains. I do the heavy work before, on the image: if the starting image is right, the prompt can stay simple.
With hybrid footage it is different, because there I use other techniques: prompts with timestamps, precise references to follow. But even there, there is no secret formula.
When the model gets it wrong
When a video model gets it wrong, it gets it wrong with a good prompt and with a bad one. You need a deep knowledge of prompting, that much is true. What you do not need are pre-packaged solutions: companies that sell you a ready-made prompt to copy and paste. It does not work, and it is not fair, because it takes people for a ride.
How we work with AI, step by step, is explained on the AI video production page. The Breil Velour case shows what happens when the work goes into the image and the material.