Ninety percent of bad results come from the source photo, not the prompt. The model can only keep what it can see, so a dim, filtered, half-cropped selfie gives it permission to invent a face. Spend two minutes here and the render actually looks like you.
Open ChatGPT, attach your photo, and paste this. Image generation is included on the free plan with a daily cap, so if you hit the limit you either wait for it to reset or run it on a paid plan.
Here is a photo of me. Generate an image of a healthier, more put-together version of the same person, keeping my face exactly as it is: same bone structure, same eye shape and spacing, same nose, same jawline, same skin tone, same hairline.
Change only the things that a few months of decent sleep, movement, grooming and styling would change. Clearer, more hydrated looking skin with texture still visible. Posture corrected, shoulders back, chin level, weight evenly on both feet. A fresh, well cut version of my current hair rather than a completely different style. Groomed brows. A simple, well fitted outfit in colours that suit my complexion.
Shoot it like a real photograph, not a beauty filter: soft even daylight, plain neutral background, relaxed natural expression, slight smile. Do not slim or reshape my face or body, do not lighten my skin, do not change my age, and do not smooth my skin into plastic. Keep it realistic enough that someone who knows me would still say that is me.The most common failure is a good-looking stranger. When that happens, do not re-run the same prompt hoping for better luck. Send this instead, in the same chat, so it still has your photo in context.
That render changed my face, so try again and treat my face as fixed. Match the photo I uploaded exactly on eye shape and spacing, nose shape and width, lip shape, jawline, chin, hairline, skin tone and undertone, and any freckles or marks you can see.
Only hair styling, clothing, posture, lighting and background are allowed to change. Before you generate anything, write me one short paragraph describing what you are keeping identical, so I can check you actually read the photo. Then generate the image.Two more things that reliably help: attach two or three photos of yourself from different angles instead of one, and iterate in small steps ("keep everything, change only the hair") rather than rewriting the whole prompt each time. If it still drifts, a stronger image model handles likeness better. Reve is the one I reach for when a render has to actually look like the person, and it is paid rather than free, so treat it as the upgrade step rather than the starting point.
Same photo, different question. Each one answers something you would otherwise pay someone to tell you.
The haircut test, so you stop screenshotting styles that will not work on your hair:
Using my photo, generate three separate images of me with three different haircuts that would genuinely suit my face shape, hair texture and hairline. Keep my face identical in all three and change nothing else: same lighting, same plain background, same top.
For each one, write two lines underneath: what the cut is actually called so I can ask for it in a salon, and how much upkeep it needs between appointments. Rule out anything that only works with daily heat styling if my hair texture will not hold it.The wardrobe test, for the version of you that owns clothes that fit:
Here is a full body photo of me. Generate four images of me in four different outfits that suit my proportions and colouring, for these settings: everyday errands, work or study, a nice dinner, and something more dressed up. Keep my face and body exactly as they are and change only the clothes, shoes and posture.
Under each image, list the individual pieces in plain language so I could search for them, name the colours that are doing the work, and tell me which single piece I could buy first that would work across all four outfits.The lock screen version, which is the whole point of the exercise:
Take the glow-up render you just made and give me a version sized as a vertical phone wallpaper, 9 by 16, with my face and shoulders in the upper two thirds and calm empty space at the bottom so my clock and notifications do not sit on my face.
Keep the same person, the same lighting and the same styling as the render, just recomposed. Do not add any text, quotes or graphics to the image.An image on your lock screen is a reminder, not a plan. This is the step most people skip, and it is the one that makes the whole thing worth doing. Run it in Claude or ChatGPT, whichever you already have open.
I generated an AI image of a more put-together version of myself and I want to turn it into something real rather than something I stare at. Here is what is actually different in the render compared to my current photo: [list the differences, for example posture, haircut, skin looking calmer, clothes that fit].
Build me a 30 day plan in three lanes. Lane one: things that are free and start today, like posture, a consistent sleep window, and how I stand and angle myself in photos. Lane two: things under [your budget], like a haircut, brow shaping, two or three wardrobe pieces, and a basic skincare routine. Lane three: the things that honestly take longer than 30 days, so I stop expecting them in a fortnight.
For every item give me the specific action, roughly what it costs, and how long it realistically takes to show. Then tell me the three things that would make the biggest visible difference first, and keep the tone kind and matter of fact rather than motivational.Used properly it is a mood board with your face on it: haircut, posture, colours, fit. Used badly it is a machine for feeling worse about a face that was never the problem. Keep it in the first category. 🌱
ChatGPT
Content Credentials, verify an image
Reve, the stronger paid image model for likeness