What You Will Finish
Before you open a generator, describe the idea in plain language: who the video is for, why it is worth watching, what happens, where the scene ends, and what must survive the process unchanged.
A beautiful idea is not yet a production task. A model can make an impressive fragment, but it will not automatically guess your direction, joke, turn, or ending. Ask it to invent everything at once and the generations start living their own lives: your eye is pleased, while the video still has no reason to exist.
Here, a model is the AI tool that follows your instruction. A prompt is the text instruction itself. Tools and prompts come later. First decide what you want the viewer to see.
Five Decisions Before the Prompt
Start with a short idea card. It does not need to sound polished or clever. Its job is to stop the video from becoming vague once you move into references and tests.
1. Who is the video for?
2. Why is it worth watching?
3. What happens?
4. What turn, result, or new fact ends the scene?
5. What must not be damaged, and what must not appear?Duration, tone, platform, and format help when you already know them. They do not replace the story, and they should not prevent you from stating its simple core.
Weak:
Make a cinematic video about space.Stronger:
A short comedy scene for viewers who recognise the beautiful promises in advertising.
A crew lands on a dangerous water planet, and a service robot mistakes it for a bargain resort.
At the end, a huge wave proves how badly the robot understood the assignment.The second version already has a viewer, expectation, mistake, and result. You can now discuss duration and production without asking the model to invent the meaning for you.
A Beautiful Frame and a Useful Frame
A beautiful frame can still fail to do the scene’s job. A close portrait of an astronaut may look expensive, yet explain nothing about the crew, ship, and wave when those are the things the viewer needs to understand. The beauty stays; the story disappears.
Treat every future frame as one line in a conversation. It should either show an event, preserve an important image, reveal the turn, or give the viewer necessary information.
Filled Example
This is one filled idea card. It is not the only valid format; it shows how the five decisions can join into a readable intent.
Who it is for:
viewers of a short comedic sci-fi video.
Why watch:
a serious expedition gradually turns into an absurd vacation advertisement.
What happens:
astronauts and a service robot arrive on a dark water planet;
the robot confidently announces that it found a bargain resort.
Turn and result:
a huge wave rises on the horizon;
the viewer realises that “ocean view” was dangerously literal.
What must not be damaged:
a dark water environment with no land;
a practical, non-toy robot;
serious delivery right up to the joke;
readable scale for the wave.
What must not appear:
colourful cartoon styling, random buildings, or an ending with no new comic fact.The card is not valuable as paperwork. After an hour of beautiful tests, it still remembers why the video exists.
Two Early Mistakes
Starting with the Model Name
“Which model is best?” tells you nothing about the task. Define the result first; choose a suitable way to produce it second. Otherwise, you are choosing a screwdriver before noticing that you are about to bake bread.
Leaving the Ending Undefined
“The character gets into a strange situation” does not explain what the viewer should understand at the end. Name the new fact, turn, decision, or visible outcome. Otherwise, the scene can end beautifully and still end with nothing. Emptiness may be minimalist, but it is difficult to edit into a story.
Practice and Check
Take one of your own ideas and fill in the five decisions. Do not write the prompt yet: there is nothing solid for it to attach to.
If one line remains vague, reduce it to one clear sentence:
I am making this video for [viewer] so they receive [promise];
the scene shows [action] and ends with [result];
I must preserve [element] and prevent [failure].Takeaway: you can explain the idea without naming a model or writing a long prompt. That is enough foundation to lay out the story in time, choose references, and decide which frames are actually needed.