The words "8K masterpiece" are not a plan. Here is a simple framework that is.
I have spent 17 years in production and VFX, and a few of the last ones writing prompts every day.
There is a common belief about AI images.
That prompting is about finding the magic words.
Add "cinematic." Add "8K." Add "masterpiece." Add "ultra detailed." Hit Generate.
Sometimes it works.
Often it does not.
The problem is that we treat an image model like a search engine. It is not one.
When you write a prompt, you are not searching for a picture. You are giving instructions for one to be built.
How a model reads your prompt
In simple terms, a model turns your words into a set of signals about what should be in the frame and how it should look.
Words near the start, concrete nouns and clear relationships usually carry more weight than vague praise.
"Masterpiece" tells it very little.
"A fisherman mending a net on a wooden jetty at dawn" tells it a lot.
Specific beats impressive.
The layers of a directed prompt
Think of a prompt like a shot brief on set. Work through it in layers.
1. Subject
Who or what is this about? Age, build, wardrobe, defining features. Be concrete.
2. Action
What are they doing, and why? A person "standing" gives you a mannequin. A person "checking the time while waiting for someone who is late" gives you a moment.
3. Environment
Where are we? Time of day, weather, era, the details that make the place real.
4. Composition
Where is the subject in frame? Centred, rule of thirds, foreground objects, negative space, what is cut off.
5. Camera
Shot size, angle and lens feel. A low angle wide lens and a long lens close-up tell two different stories about the same face.
6. Lighting
Where is the light coming from, and what colour is it? A window on the left, a neon sign behind, a single overhead bulb. Give it a source.
7. Mood
What should the viewer feel? Use it to guide the other layers, not as the only instruction.
8. Style
Photographic, illustrated, film stock look, a colour palette. Keep it consistent across a series.
9. Constraints
What must not happen. No text in frame. Hands visible and natural. Keep the logo unchanged. Constraints save more takes than adjectives.
An example, before and after
Before:
A cinematic portrait of an old man, 8K, masterpiece, dramatic.
After:
A 70-year-old fisherman with a grey stubble beard and a faded blue shirt, mending a torn net on a wooden jetty. Early dawn, soft warm light from the left, mist on the water behind him. Medium close-up, eye level, subject on the right third, shallow depth of field. Calm, tired, proud. Natural photographic look. No text in frame.
The second prompt is longer, but every word is doing a job.
Change one thing at a time
When a result is close but not right, do not rewrite the whole prompt.
Change one layer.
Swap the lens. Move the light. Shift the time of day.
That is how a cinematographer works on set, and it is how you learn what actually moves the image.
Iteration with intent is cheaper than iteration by luck.
Use references when words run out
Some things are hard to describe. A specific face, a costume, a location you have already built.
That is where reference images help. They lock what words cannot.
For a series or a film, consistent references matter more than clever wording. Without them, your character slowly turns into someone else.
Prompting for a film, not a single image
One great image is a poster.
A film needs the same character, the same world and the same light logic across dozens of frames.
So keep a shot brief for each frame and reuse the layers that must stay the same: subject, wardrobe, location, palette.
Only change what the shot needs to change.
Where FrameOn fits
We built FrameOn as an AI filmmaking platform for filmmakers, studios and brands, and a lot of it is about taking this structure off your shoulders.
You can lock characters and locations once and reuse them across every shot.
Cue, the production agent in the app, helps turn a story into a script and a shot list, so each frame starts with intent.
And every shot is priced before it runs, so you decide which takes are worth spending on.
Direction is the prompt
The best prompt is not the longest or the most decorated.
It is the one that carries a clear decision.
Know what you want, say it in the language of the shot, and protect what must not change.
FrameOn's early access is open at frameon.us.
— Founder, FrameOn
FrameOn Creativity is yours. The tools are ours. frameon.us
