Here’s how to make the shot you didn’t have time to take… and still look exactly like yourself in it. Your face, your hair, your skin, your spaces that weren’t quite right. The semi-real version is the only one worth making.
What you get: one prompt you can reuse forever, two tools to run it in, and the photos you meant to take and didn’t. Still recognizably you.
Why I started doing this
I was failing at posting consistently. Not “struggling.” Failing. And not having a photo ready was one of the excuses that stopped me.
So I started generating them. And immediately felt like I was cheating.
The fix was keeping myself in them. Double chin. Backrolls. Clothes I own. Rooms I stand in. A semi-real me, not an avatar.
The second it stopped being a fantasy version of me, it stopped feeling like cheating. And I started showing up.
The most important rule
Every image model’s factory setting is to correct you.
Clearer skin. Thicker hair. Straighter teeth. A few years younger. A thigh gap you never asked for and abs nobody believes. It doesn’t matter what you look like… it will find something to fix, and whatever it fixes is precisely the thing that makes the image stop being you.
People who know you will feel it before they can name it.
Say this in every prompt:
Keep the person exactly as she appears in the reference images. Same face, same hair, same skin, same body, and the same tattoos in the same places, if any. Do not beautify, smooth, slim, de-age, or correct anything. Keep natural skin texture, pores, and any blemishes.
Pick your tool
Nano Banana (free). Start here today. Google’s image model, at gemini.google.com/images. Free, no card. Upload 3 to 5 photos of yourself, paste the prompt, go. One image at a time, and it doesn’t remember you, so you re-upload your photos every session. Outputs carry a visible watermark plus Google’s invisible one.
OpenArt ($14/mo). What I use. openart.ai/suite. Start on the free tier, but you’ll run out fast. Build a Character trained on your photos once, and it knows you every time after. No re-uploading. Up to 4 images at a time. This is the one my actual content comes out of.
Your ten-minute reference shoot
Do this once and you’re set for months. Prop your phone against something in front of a bright window, natural light only, no flash. Wear the makeup you wear on a normal day, not full glam. These do not need to be post-ready shots. They need to be accurate ones.
- Front facing, smiling with teeth
- Front facing, smiling without teeth
- Looking off into the distance
- Laughing
- Serious, no smile
- Neutral, resting face
- Suspicious, one eyebrow up
- Hair up the way you wear it up (a messy bun absolutely counts)
- Full body, head to toe
- A shot that shows your tattoos, in something you’d normally wear
- Wearing your usual jewelry
Always upload 3 to 5 photos of yourself, not one. Different angles. If you have tattoos, piercings, glasses, freckles, or a scar you want kept, make sure at least one photo shows them clearly. It can only keep what it can see.
The prompt that writes your prompts
Copy the prompt below and paste it into Claude, ChatGPT, or Gemini, whichever one you already use. I use ChatGPT for this because it’s cheap and it describes pictures better than the others.
Then send it your inspiration photo. It writes you a paragraph describing that photo’s outfit, setting, lighting, and camera angle, and says nothing about the person in it.
You paste that paragraph into your image tool along with your own photos. Your photos tell it what you look like. The paragraph tells it everything else.
You're my image prompt writer.
I'm going to send you an inspiration photo — something I saved off Pinterest, someone's Instagram, or a shopping site. I do not want the person in it. I want the scene, with me in it instead.
So here's the job: write me one prompt that recreates everything in that photo EXCEPT the person's appearance. I'll upload my own photos alongside your prompt, and the image tool will use those as the reference for who she is. Your paragraph builds the world around her.
THE ONE HARD RULE — never describe her appearance. No age, no face, no body type, no height, no weight, no build, no skin tone, no eye color, no hair color, no hair length, no hair texture, no ethnicity. Not one word. Every detail you add about how she looks overrides my reference photos and the output stops being me. Refer to her only as "the woman in the reference images."
DO describe what she's doing. Pose, body position, where her weight sits, what her hands are doing, where she's looking, and the expression as an action — mid-laugh, caught mid-sentence, glancing off camera, eyes closed.
DO describe the styling. The outfit in real detail: garment, fabric, color, cut, how it sits and moves, wrinkles, a sleeve pushed up, a strap slipping, a hem catching. Shoes, jewelry, bag, sunglasses, nails, watch, whatever's on her. Hair STYLING only — messy bun, half-up, tucked behind one ear, air-dried, wet from the ocean — never its length, color, or texture.
DO describe the place. Real depth, and enough detail that it feels lived in rather than staged. Name at least three specific objects that make it somewhere a person actually spends time: a mug on the counter, a chair pushed out, a jacket over the back of it, a car in the driveway. Empty perfect backgrounds are the fastest way to look fake.
DO describe the light. iPhone flash, golden hour through a window, flat overcast afternoon, kitchen overheads, lamp light. Say where it comes from and what it does to the shadows.
DO describe the camera. Angle, distance, how much of her is in frame, negative space, what's cropped out. Pull the camera back — no tight crops unless I ask. Name the lens feel: phone-wide, slight distortion up close, shot from slightly below, over a shoulder.
DO describe the photo's imperfections. Slight motion blur, an off-center horizon, a shadow falling across part of the frame, the half-second before or after the pose. It should read like a person took it, not a crew.
Format: one paragraph, no bullets, no headers, no labels. Vertical 4:5 unless I say otherwise. Written so I can grab it in one copy.
End every prompt with this exact line: "Keep the woman exactly as she appears in the reference images — same face, same hair, same skin, same body, and same exact tattoos in the same exact locations on the body, if any. Do not beautify, smooth, slim, de-age, or correct anything about her. Keep her natural skin texture, pores, and any blemishes. This should look like an unedited phone photo, including its flaws."
You can improve the composition if it makes the shot stronger — better spacing, more room around her, a cleaner angle. Never change her.Save it somewhere you’ll find it. Your notes app, a Google Doc, or bookmark this page.
If you don’t have an inspiration photo
Skip the screenshot and describe the scene yourself, in normal words. This is what I type most days:
“Take the woman in the reference images. Put her in an office with a cup of coffee and her laptop, laughing at the screen on a Zoom call. Hair up in a messy bun.”
Same rule either way. Describe the place, the clothes, and what she’s doing. Never describe her face, hair, or body.
How to change the image shape
The prompt makes tall 4:5 images by default. To get a different shape, change that one line near the end of the prompt.
| Ask for | You get | Use it for |
|---|---|---|
| 4:5 | A tall rectangle | Carousels and feed posts |
| 9:16 | A full phone screen | Reels, Stories, TikTok |
| 1:1 | A square | Your profile grid, thumbnails |
| 16:9 | A wide rectangle | YouTube, websites, slides |
How to do it, five steps
You do the photo shoot once, then repeat steps two through five forever.
- Pick your best three to five from the shoot. Different angles, and at least one showing your tattoos, glasses, piercings, or freckles if you have them. This is the ingredient you can’t fake.
- Go find a photo you love. Pinterest, someone’s feed, a shopping site, a stock photo. Screenshot it. You’re stealing the scene, not the person. Pick a real photograph, not an AI one.
- Hand the screenshot to the prompt writer. Paste the prompt into whichever AI you already use, drop in your screenshot, get a paragraph back. Skim it once and edit anything that doesn’t match your vision.
- Open your image tool. Upload your photos, paste the paragraph, generate.
- Fix the one thing that’s wrong. It’s always one thing. Say it plainly: “her hair looks too short, match it to my reference photos.” Regenerate. Done.
Don’t use an AI image as your inspiration. Pinterest is flooded with them now and they’re seductive… perfect light, perfect skin, perfect everything. Reverse-engineer an already-too-polished image and your output inherits all of it. How to spot one fast: hands and fingers, text on signs, jewelry that melts into skin, hair with no flyaways, and a background where nothing is worn, scuffed, or plugged in.
The three things that make it look fake
- The person got over-corrected. Skin smoothed, hair thickened, teeth straightened, a few years shaved off. Fix: name the reference photos as the source of truth and forbid the correction outright.
- The room is too clean. Empty counters, perfect walls, nothing plugged in, nothing out of place. Fix: name three specific real objects in the background.
- The camera is too close. Tight crops read as portrait mode, and portrait mode reads as generated. Fix: pull back.
The bottom line
You’re not making a better-looking you. You’re removing the thing that stops you from posting.
You skip posting because there’s no time. Because the light is wrong, you’re not dressed for it, the kitchen behind you is a disaster, and you’d need an hour you don’t have.
This takes all of that off the table. You still show up as yourself. You just stop waiting for a perfect day to do it.
Use these for carousel backgrounds, reel covers, story slides, the “here’s me working on this” shot you didn’t have time to take.
This is the kind of thing I teach live inside the AI Skool of Business… practical AI for founders who want to show up consistently without it eating their week. Just you, me and AI.