Twelve prompts, one method. AI Central's photorealistic Midjourney set chases realism with none of the usual equipment, no parameter flags, no camera jargon, no version numbers. Every prompt is written as a flat description of a photograph that already exists, subject first, then its position in the frame, then wardrobe, expression, background, lighting, composition, perspective and style. The word photorealistic never appears inside a single one of them. Description density does the work, not vocabulary.
Reviewed August 7, 2026, against the source document's full text.
What the document actually contains
The deck runs twelve prompts, each labeled with a type. Three are product shots. Two are fashion and two are portraits. The remaining five are one each of commercial, lifestyle, interior design, editorial and architecture. Eight distinct categories across twelve slots, which tells you the set was built to cover a spread of commercial briefs rather than drill a single look.
The prompts themselves run between 131 and 188 words, averaging 153. That is long by the standard of most prompt libraries, and the length is the whole point.
The prompts describe, they do not instruct
Not one of the twelve opens with a verb. There is no create, no generate, no make me an image of. Every prompt opens by naming a subject and putting it somewhere in the frame, in flat declarative present tense, the way a caption writer describes a photo already on the desk in front of them.
The editorial prompt, number ten, goes as far as recording what cannot be determined.
The subject's ethnicity is indeterminate.
That line is AI Central's, from the tenth prompt in the deck. It is a strange thing to tell an image model. You are not describing someone you invented, you are hedging about someone you are looking at. The same tic runs through eight of the twelve, in phrases like likely Black, likely in his twenties or thirties, and a car that appears to be a classic or vintage model.
The document does not say how the prompts were built, so what follows is inference rather than fact. That hedging is the fingerprint of description, not invention. Someone was looking at reference images and writing down what they saw, uncertainty included, and the uncertainty survived into the prompt.
It turns out not to hurt. A model handed likely Black still produces a specific person. What the phrasing gains is the register of an observation instead of a specification, and observation is what produces photographs.
The slot order that repeats twelve times
Read the twelve in sequence and the same skeleton surfaces. The prompts move through a fixed set of slots, in roughly this order.
- Subject, named in the first clause, with age, build or material where it applies.
- Position in the frame, stated outright, centrally positioned, slightly off-center to the left, in the mid-ground.
- Detail on the subject, clothing, texture, color, accessories.
- Expression or state, for anything living.
- Background, usually stripped back to a plain color.
- Lighting, always in a sentence of its own.
- Composition, what the frame emphasizes.
- Perspective, where the viewer sits relative to the subject.
- Style and atmosphere, the closing verdict on how the image should feel.
This is not a loose tendency, it holds. All twelve prompts name the lighting. Eleven of twelve name the composition. Eleven place the subject in the frame explicitly. Ten name the style. Nine use the word perspective outright. The handful of gaps are cosmetic. Prompt seven never uses the word composition, and still tells you the bottle is the central focus with the hat at the lower left.
What is deliberately absent
Search all twelve for Midjourney's own syntax and you find nothing. No aspect ratio flag, no version flag, no stylize or chaos value, no double colon weights, no niji. Search for camera language and you come up empty too. No focal length, no aperture, no film stock, no render engine, no 8k, no bokeh, no depth of field.
Most striking of all, the word photorealistic appears in the deck's title and never once inside a prompt. Neither does realistic, photograph or camera.
That absence is the actual technique. The common failure in image prompting is to reach for realism as an adjective, stack photorealistic and hyperrealistic and 8k, then wonder why the output looks like a video game cover. Words about realism behave as style tokens, and style tokens pull toward the aesthetic of whatever they were attached to in training. Describing what is physically in the frame, in order, does not.
Closing the frame instead of negating it
There is a second move worth stealing. Four of the twelve end by stating what is not there, phrased positively, as a plain fact about the scene. The commercial prompt, the silver Porsche shot from overhead, closes like this.
There are no other objects or elements in the image except for the car and the floor.
That is AI Central's line, from the fifth prompt in the deck. It does the job a negative prompt would do without needing one. The same instinct shows up in the black and white portrait.
The setting is a plain, light-gray background, devoid of any specific environment.
Seven of the twelve name a background at all, and where they do it is nearly always flattened to one color, light gray, deep neutral gray, dark gray, light beige, cream. The empty background is not laziness. It is a decision to spend every available word of detail on the subject, which is what a studio photographer does with a seamless backdrop.
How these counts were made
A method note, because the numbers matter. I took the document's transcript, split it at each prompt marker, and isolated the twelve prompt bodies from the surrounding deck furniture, the running title, page headers and repeated download line. The category counts come from the type labels the deck prints above each prompt. The absence checks were literal string searches across the combined prompt text. One caution, the transcript is the text layer of a PDF, so anything baked into an image would not appear in it. The twelve bodies came through intact.
What to do with this
If you have never written a long prompt, do not start by inventing one. Start by describing one.
- Find a photograph you want to imitate, one you already have or one you can picture clearly.
- Write the subject in the first sentence, and say where in the frame it sits.
- Add one sentence each for background, lighting, composition and viewing angle. Four sentences, no exceptions, even when the honest answer is plain gray and flat light.
- Close with what the image does not contain, if the frame is meant to be empty.
- Do not add the word photorealistic. Spend that space on another sentence of description instead.
If you already use Midjourney daily, the useful experiment is subtractive. Take a prompt you have been feeding parameter flags, strip them out, and spend the words you save on lighting and perspective. These twelve work as a control group for that test. They are all long, all flagless, and all built the same way, so any difference you see is coming from the description.
Where the approach runs out
Two honest limits. The first is cost in effort. A hundred and fifty words per image is fine for a hero shot and miserable across a batch of forty. The second is durability. Nothing in the document addresses model versions, and prompts written against one release of an image model rarely behave identically on the next. Treat the slot ordering as durable and the exact wording as disposable.
A few questions that come up with this set.
Do these prompts only work in Midjourney
The deck is titled for Midjourney and offers the twelve as Midjourney experiments, so that is what the document speaks to. Because none carry any Midjourney syntax, though, there is nothing in the text that would break if you pasted it into a different text to image tool.
Why is there no aspect ratio or version flag anywhere
The document does not explain the omission, it simply never uses them. What the text shows is that all twelve are built to work on description alone, which makes them easy to lift into a different tool or a newer model version without rewriting.
Do I really have to write a hundred and fifty words every time
That is the range the deck works in, 131 words at the shortest and 188 at the longest. If you want a shorter version, cut detail on the subject before you cut the four structural sentences, background, lighting, composition and angle. Those four are what the entire set holds constant.
Can I keep one prompt and just swap the subject
That is close to how the set is built. Prompts one and twelve are both labeled fashion, both place a single figure against a plain flat backdrop under undramatic light, and the difference between them lives almost entirely in the subject sentences. Holding background, lighting, composition and perspective steady while changing the subject is a sound way to keep a look consistent across a series.