How to Write Prompts From a Reference Photo

AI PHOTO PROMPTSREFERENCE PHOTO PROMPTS8 COPY-PASTE PROMPTS

Scrolling Instagram, you stop on one photo. Warm window light, a quiet face, a muted green wall. You want that feeling in your own picture, but 'make it look like this' never works. A reference photo prompt fixes that: you study the image, put its building blocks into words, and then write an original prompt with your own subject and setting.

The important part is the word original. We are borrowing ideas like light direction, colour palette and framing, which are not owned by anyone. We are not copying someone's face, a named person's likeness, or a brand's artwork. Use photos you took yourself or that you have the right to study, and attach only your own face for the person in the result.

Below you get a simple breakdown method and eight templates for common situations, such as analysing light, pose, colour and background. Results vary, so expect a bit of experimenting.

Copy-Paste Prompts: Reference Photo Prompt (8 Ready Ideas)

Look at the attached reference image. Describe it in seven short labelled lines: SUBJECT pose and framing, OUTFIT style (colours and fabric only), SETTING, LIGHT direction and softness, CAMERA lens feel and angle, STYLE colour mood, and mistakes to AVOID. Do not name any real person or brand. Keep each line under 20 words.

Study the lighting in the attached reference photo. Tell me: where the main light comes from, whether it is soft or hard, the colour temperature (warm or cool), where the shadows fall, and whether there is a fill or rim light. Then write one sentence I can paste into a prompt to recreate only that lighting on a different subject.

From the attached reference photo, list the five main colours as simple colour names (for example dusty teal, warm cream, burnt orange). Say which colour dominates, which is the accent, and how saturated the overall look is. Then write a STYLE line I can reuse that describes this colour mood without naming the original photo.

Describe only the pose and framing in the attached reference photo: body angle, head tilt, where the hands are, how much of the body is visible, and where the subject sits in the frame (centre, left third, right third). Write it as a SUBJECT line using neutral words, with no description of the person's face or identity.

SUBJECT: Use my attached photo, face unchanged, [pose from breakdown]
OUTFIT: [colour and fabric from breakdown], my own style
SETTING: [setting idea, new location of my choice]
LIGHT: [light sentence from analysis]
CAMERA: [lens and angle from breakdown]
STYLE: [colour mood line from breakdown]
AVOID: copied backgrounds, logos, text, extra people

Using the style notes below, create a new image that shares only the mood, not the content. Notes: [paste your light, colour and camera lines]. Use my attached photo with my face unchanged. Choose a completely different location, outfit design and pose from the reference. Original composition, realistic photograph, no text.

Looking at the attached reference photo, describe only the background in a way I can reuse: type of place, three key objects, depth and blur level, and colour tones. Do not include the foreground subject. Then rewrite it as one SETTING line for my own portrait, changing at least two objects so it is not an exact copy.

I attached a reference photo and my generated result. Compare them. List three differences in light, colour and composition that make my result look less natural. Then rewrite my prompt below to fix those three points while keeping my face unchanged. Explain each fix in one plain sentence so I learn from it. My prompt: [paste your prompt here].

How do I break a photo into five parts?

Whenever you analyse a picture, look at five things in the same order: light, pose and framing, colour, background and camera feel. Write one short line for each. That gives you the same labelled structure we use in our other prompts, so everything snaps together later.

Do it with a pen on paper if you like. Seeing 'light from the left, cream and olive palette, waist-up' written plainly makes it obvious which parts you want to keep and which parts you will swap for your own balcony, your own kurta, your own city.

• Light: direction, softness, warm or cool
• Pose and framing: angle, hands, crop
• Colour: dominant colour, accent, saturation
• Background: place, objects, blur
• Camera feel: lens length, height, angle

Where do I paste the reference, and where my own photo?

Step one: upload the reference image alone and ask for the breakdown. Step two: read it, fix anything wrong, and delete anything that identifies a person or brand. Step three: open a fresh chat, attach your own photo, and paste the assembled prompt. This works the same way in Gemini and in ChatGPT.

Keeping these as two separate steps works better than uploading both images at once, because the tool is less likely to mix the reference person's features into your face. If you notice the result looks like the reference model instead of you, restart and add 'face from my uploaded photo only'.

How do I borrow a mood without copying someone?

Borrow the mood, not the person. Do not ask the tool to recreate a celebrity, a named individual, a copyrighted character or a brand's ad. If you notice the result looks too close to the source, change the setting, outfit and pose and try again.

When you recreate photo style with ai, aim for 'inspired by the feeling', not 'identical copy'. This is better ethically and usually looks better too.

What does a good breakdown actually look like?

A good breakdown is specific enough to repeat and short enough to edit. Suppose your reference shows a woman reading near a window. A weak breakdown says 'pretty light, nice mood'. A strong one says 'soft daylight from the left at about 45 degrees, warm white tone, gentle shadow on the right cheek, muted green and cream palette, shot from slightly above at around 50mm'. You can paste the second one into a prompt and test it.

Check that every line answers a question: where does the light come from, what is the main colour, how close is the camera? If a line cannot be tested in a generated image, delete it. Words like 'aesthetic' or 'vibes' give the model nothing to hold.

• Each line names one measurable thing.
• Colours are named in plain words.
• Direction words such as left, right, above and behind are included.
• No person, brand or artwork is named.
• The whole breakdown fits in under 120 words.

Can a cafe window photo become my own balcony portrait?

Imagine you love a picture of someone at a cafe window with warm afternoon light. Run the first template and you might get: light from the left, warm and soft; framing waist-up, subject on the right third; palette of cream, brown and olive; background of blurred shelves; camera 50mm at eye level.

Now make it yours. Keep the light line, the framing and the palette, since those are the mood. Change the place to your own balcony with potted plants, change the outfit to a different style, and change the pose slightly: if the original looked down at a cup, yours looks out toward the street. That is three changes, and the result is clearly your picture.

Paste everything into the 'Recreate With My Own Photo' template with your own face attached. If the result feels too close, run the 'Make It Different' template. This is a safe way to recreate photo style with ai without copying anyone.

Why does the AI describe my reference photo wrongly?

AI tools describe pictures confidently but not always correctly. A common slip is the light direction, which the model may flip left and right. Look at the shadows yourself before trusting it. Another slip is guessing details that are not visible, like a camera model or the age of a person; delete these lines.

A third problem is over-long output. A reverse prompt from image result with two hundred words will confuse your generation step. Ask for 'seven lines, under 20 words each'. If the result still does not match, run the comparison template and let the tool tell you what is different. Treat the description as a draft, not a final answer.

Which small habits make the breakdown better or worse?

Ask the AI to describe only what it can see, not to guess. If it says 'shot on a specific camera', ignore it and keep the lens feel instead. Use plain language and short lines. If the description runs long, ask for 'a version under 70 words', because short prompts are easier to edit and test.

Watch for three slips. Pasting the long description as-is: trim it to the seven lines. Forgetting the face line: always start the SUBJECT with 'use my uploaded photo, face unchanged'. Keeping the same background and outfit as the reference: change at least two things, so the result is clearly your own idea and not a copy in a different face.

Frequently Asked Questions

Final Thoughts

A good reference photo prompt turns a vague 'I want that vibe' into clear words: light, pose, colour, background and camera. Study the picture, keep what is useful, change what makes it yours, and always use your own face photo.

Pick one photo of your own that you love, run the light analysis template first, and build outward from there. Once you are comfortable, compare your breakdowns with other portrait and festival prompts and notice which lines they share.

Note: AI image and text results vary by tool, version and settings. Use only your own photos, avoid copying real people’s likeness, and follow the terms of the AI tool you use.