Ninety-Nine Image Prompts You Will Not Use: The Five Slots That Actually Matter
Every few weeks another pack appears: ninety-nine viral image prompts, a thousand commands, copy and paste and get stunning art.
You will use about four of them. That is not a criticism of the pack – it is how prompt packs work, and knowing it in advance changes what you take from one.
Because the useful part was never the ninety-nine lines. It is the structure underneath them, which is the same five slots every time, and which you can write yourself once you can see it.
The five slots of an image prompt
Take any prompt from any pack and pull it apart. It will be these, in some order, with the weak ones leaving two or three empty.
| 1. Subject → | 2. Style → | 3. Light → | 4. Framing → | 5. Constraints |
|---|---|---|---|---|
| What is in the picture
One clear thing doing one clear action. Two subjects is where most images fall apart. |
How it is rendered
Photograph, illustration, 3D, paper cut. Naming a medium beats naming an artist. |
Where the light comes from
The slot that separates amateur output from professional-looking output, and the one people skip. |
Shot and composition
Close-up, wide, eye level, negative space on the right. This is where thumbnails are won. |
What to leave out
No text, no extra hands, no clutter. The slot that saves the image. |
Repeatability is the real prize. A single striking image is luck; a prompt whose slots you can adjust deliberately is a thumbnail process. That is the same argument as the five slots for text prompts in vague in, vague out – specify the job, and the tool stops guessing.
Five rewrites, for jobs a channel actually has
The same five jobs come up over and over. Here is the version most people type, and the version that produces something you can publish.
| The job | What most people type | The rewrite |
|---|---|---|
| Thumbnail subject | “A shocked man looking at a laptop” | “Close-up photograph of a man reacting to something off-screen, hard side light from the left, dark background, large empty space on the right for text, no text in the image” |
| B-roll you cannot film | “A busy office” | “Wide shot, empty open-plan office at night, overhead strip lights, one monitor glowing, slight haze, no people, no logos” |
| Background plate | “A nice gradient background” | “Flat dark navy background with a soft radial glow in the lower left, no objects, no pattern, room for a cut-out subject on the right” |
| Explainer graphic | “An illustration about money” | “Simple paper-cut illustration of a single coin on a flat surface, top-down, soft shadow, two colours only, plenty of white space” |
| Thumbnail with words in it | “A thumbnail that says FREE” | “Bold block lettering reading FREE, centred, high contrast, no other text, no extra characters” – then check every letter. |
Look at how often “no” appears. Negative constraints do more work in image prompts than in text ones, because the model will happily add a second hand, a watermark-looking mark, or a caption you never asked for. Naming what you do not want is half of the instruction.
Watch the reel
The prompt set this article was written around. Posted on the BANI Academy page.
What a prompt pack actually gives you
Being straight about this, because packs get sold as results and they are not.
A pack is somebody else’s taste, written down. The prompts produce the look the author likes, for the subjects the author cares about. That is genuinely useful as a starting vocabulary – you learn that “hard side light” and “shallow depth of field” are things you are allowed to ask for – and useless as a shortcut, because your channel does not look like their portfolio.
The numbers are marketing. Ninety-nine, a thousand, five thousand. The count has almost no relationship to how many you will use, and a pack of twelve prompts you actually understand beats a thousand you scroll past. We counted the lines on one of these lists in the hundred-prompt audit and found it was not even the number on the cover.
They go stale differently from text prompts. Image models change behaviour between versions more visibly than language models do. A prompt tuned to one release can produce something noticeably different on the next, which is another reason to learn the slots rather than memorise the lines.
Thumbnails are a different problem from art
Almost every image-prompt pack optimises for a picture that looks impressive at full size. A thumbnail has to survive being two centimetres wide on a phone, next to eleven others, in a fraction of a second.
| What thumbnails need | What “stunning AI art” usually delivers |
|---|---|
| One subject, instantly readable | A rich, detailed scene that turns to mush when shrunk |
| Hard contrast between subject and background | Beautiful, even, low-contrast lighting |
| Deliberate empty space for your text | A composition filled edge to edge |
| A face or an object doing something unresolved | A finished, balanced, resolved image |
| The same look repeated across videos | A different style every generation |
That fourth row is the one worth sitting with. A resolved image has nothing left to ask, and a thumbnail exists to raise a question – which is the whole mechanism we set out in thumbnail ideas that make people ask a question. Generated images default to resolved, so you have to ask for the opposite deliberately.
Two practical constraints on top. If words have to appear in the image, check every letter before you publish – generated lettering produces convincing nonsense, and a misspelling in a thumbnail is worse than no text at all. And whatever ends up on screen, platform rules on synthetic material apply regardless of how it was made: disclosing generated visuals without losing trust covers handling that without deflating the video.
The same discipline as photo work, in fact: name the repair, name the limit. We argued it for family pictures in the shoebox photo, and it applies just as neatly when you are generating rather than restoring.
Frequently asked questions
Do I need a paid pack?
No. Take five prompts from any free set, pull them apart into the five slots, and you have the pattern. That is the part that transfers.
Why does my image have text I did not ask for?
Because you did not say not to. Negative constraints are a normal part of the prompt, not an advanced technique.
Which image tool should I use?
Whichever handles your most common job. If words appear in your thumbnails, pick one that renders text legibly – that difference matters more than overall quality.
How many attempts should one thumbnail take?
Three, if the prompt is specified properly. If you are on attempt fifteen, the problem is a missing slot, not the model.
Take four prompts apart tonight
Open any pack you already have. Choose four prompts, and for each one write out which of the five slots it fills and which it leaves empty. That exercise takes about twenty minutes and it is the entire content of every pack you will ever buy.
Then write one prompt of your own for your next thumbnail, with all five slots filled, and test it at thumbnail size before you decide anything.
Learning a lot does not make anyone good. Doing a lot does.
If you want the structure that turns that into a habit, that is what our free challenge is built for: no barrier to entry, a genuine commitment to action, and a refundable commitment fee you get back when you finish. You are not paying for the knowledge. You are betting on yourself finishing, and we hold the stake.
Want the prompt set from the reel? Ask for it in the comments on the original post and we will send it over – then pull four of them apart before you use any.
For more on building channels for English-speaking audiences, that is what we work on at mmoyoutube.com.
