The Perfect AI Photo Prompt: A Five-Part Formula That Works Every Time

Most people write prompts the way they write a text message: a subject, a couple of adjectives, and hope. The model fills every gap you leave, and gaps are where the plastic skin, the impossible lighting, and the generic stock-photo look come from. A professional prompt is not longer for the sake of it, it is complete. It answers the same questions a photographer answers before pressing the shutter: who is in frame, where are they, how is it lit, what lens is on the camera, and how is the final file graded.

The Five Parts, In Order

Order matters because image models weight the opening of a prompt most heavily. Lead with the subject, then widen out to the world around it, then describe light, then the camera, then the finish. Anything that arrives late reads as a suggestion; anything up front reads as a requirement.

• Subject: who or what, plus the one detail that makes it specific ('a silver-haired chef', not 'a person').
• Environment: the surface, the background, and one prop that grounds the scene.
• Light: direction, quality and colour ('soft window light from camera left, warm').
• Camera: body, focal length and aperture, or at least focal length and depth of field.
• Finish: the grade, the grain, the render quality ('editorial colour, fine grain, tack-sharp').

The Formula: [subject with one defining detail], [environment: surface, background, one prop], [light: direction + quality + colour], shot on [camera] with a [focal length] lens at [aperture], [finish: grade, texture, sharpness], photorealistic

A Vague Prompt Versus a Complete One

Compare the two below. The first is what most people type. The second is the same idea with the five parts filled in. The subject has not changed at all, only the amount of decision-making you have taken back from the model.

Vague: a chef in a kitchen, professional photo, high quality, 4k

Complete: A silver-haired chef plating a dish, stainless prep counter with scattered herbs and a copper pan behind, soft window light from camera left with a warm bounce fill, shot on a Canon EOS R5 with an 85mm lens at f/2, editorial colour grade, fine grain, tack-sharp focus on the hands, photorealistic

Specific Beats Superlative

Words like 'beautiful', 'amazing', 'high quality' and '4k' are noise. They do not describe anything the model can render differently. Replace every superlative with a measurable detail: instead of 'dramatic lighting', write 'single hard source at 45 degrees with deep falloff'. Instead of 'high quality', name the camera and the grade. The rule of thumb is simple — if two different photographers would shoot it differently, your prompt is not specific enough yet.

Change One Variable at a Time

When a result is close but not right, resist the urge to rewrite the whole prompt. Change one part, regenerate, and compare. Swap the 85mm for a 35mm and the intimacy disappears. Swap 'soft window light' for 'hard afternoon sun' and the mood flips. Testing one variable at a time is how you build an instinct for what each term actually does, and within a week you will be reaching for the right word the first time.

• Keep a note of the prompts that worked; the good ones become templates.
• If the image is busy, remove a prop rather than adding 'simple' or 'clean'.
• Put the most important detail first — the model listens hardest to the opening words.
• Only add a negative list once something has actually gone wrong; do not pre-emptively bloat the prompt.

Fill in the five parts, test one variable at a time, and keep the winners. That is the entire method, and it works in Manus, ChatGPT and Gemini alike because it is not a trick — it is a briefing.