How I Prompt For AI Images And Video
People often ask me for prompts.
I get why. A good prompt looks like the secret. It feels copyable. It gives you the tiny thrill of, “ah, that is the spell.”
But the useful thing is not the exact sentence. The useful thing is how the prompt was built.
I Use LLMs As Drafting Partners
I use LLMs to help write prompts, but I do not let them direct the piece.
My job is to know what I want. The LLM helps me organize it. It can translate a messy idea into shot language, expand visual references, create variations, or turn a paragraph into a cleaner set of constraints.
The danger is letting the LLM add too much taste.
If every prompt becomes “cinematic, highly detailed, dramatic lighting”, you are not directing anymore. You are letting the autocomplete machine decorate your indecision.
So I use LLMs for structure:
- shot type
- subject
- action
- setting
- lighting
- camera movement
- texture
- continuity rules
- things to avoid
Then I cut the parts that sound like perfume.
Images Need Different Prompts Than Video
Image prompts can hold more detail because the frame only has to exist once.
Video prompts need to think about time. What changes? What stays still? Is the camera moving, or is the subject moving? Does the shot need a beginning, middle, and end? Should the model maintain a reference, or can it invent?
For video, I care less about poetic language and more about behavior.
Bad video prompt:
A beautiful cinematic woman in a futuristic city.
More useful video prompt:
Medium close-up of the same woman from the reference image standing under red neon rain. Camera slowly pushes in. She turns her head slightly toward camera, holds eye contact, then looks past camera as the background traffic stays soft and out of focus.
It is not prettier. It is easier to perform.
Character Sheets Help
If I need the same character more than once, I try to build a character reference.
That might be a clean portrait, a full-body image, a three-view sheet, wardrobe details, or a set of expression references. The goal is to reduce the number of things the model has to invent every time.
A character sheet does not solve everything. It gives you a better starting point.
You still need to watch for hair drift, wardrobe changes, face shape shifts, and the model slowly turning your character into a cousin with better cheekbones.
Storyboard Panels Are Better Than Vibes
Storyboard panels are one of the strongest upgrades I have seen in AI video work.
A prompt describes a shot. A storyboard shows the shot in context. It gives the model and the human a shared target: composition, action, rhythm, and transition.
For longer pieces, I like panels because they force me to answer questions early.
Where is the camera? What does the character know in this moment? What changed since the last shot? Is this a new idea, or just another pretty frame?
That last question hurts. Good.
Prompting Is Really Pre-Production
The best prompts usually come from pre-production, not clever wording.
References, shot lists, character rules, camera language, edit rhythm, and continuity notes do more work than magic adjectives. The prompt is where those decisions enter the model.
If you want help with prompting, I will not just hand you a list of phrases. I can help you build a prompt process that fits the images, videos, characters, and stories you are trying to make.