Generating Age-Progression Photos of Princess Diana Using AI
I spent about three weeks last month working on a project that involved generating realistic aged versions of Princess Diana for a documentary research team. I've since done similar work for a few other historical figure projects, so I have a sense of what works and what doesn't when you're dealing with this kind of image generation. Here's the straightforward breakdown. The core tool people reach for is Midjourney v6, paired with a face-swap or aging pipeline if they need consistency across multiple images. The phrase you'd typically search for when researching this is What Would Princess Diana Look Like Today A Modern Reflection Of Timeless Grace, and you'll find it scattered across forums, articles, and social media posts where people discuss the concept. But searching for that won't give you the actual result — it'll give you speculation. The real output comes from running controlled generations.
What Would Princess Diana Look Like Today A Modern Reflection Of Timeless Grace
The technical approach starts with finding the best reference image. You want something from the early 1990s — ideally a well-lit portrait where the facial structure is clearly visible. Images from 1995 onward tend to have heavier makeup or lighting that obscures bone structure, which makes the aging process less accurate. The National Portrait Gallery in London has high-resolution scans of several photos from that period that work well. I used a 1992 portrait of Diana in a pale blue dress for my project. Once you have the reference, you feed it into Midjourney v6 with a prompt that specifies the age range and desired outcome. Something like a realistic age-progression of the person in the image to around age 60, wearing contemporary clothing, in natural lighting, photorealistic style. The v6 model handles skin texture and aging much better than previous versions. Early models would just make the face wrinkled and generic-looking. V6 actually produces variations in skin tone, subtle asymmetry, and age-appropriate features like nasolabial folds and crow's feet. Here's where things get complicated. Face swapping is necessary because Midjourney won't replicate the exact facial features from your reference image. You generate the base image first, then use a tool like InsightFace or ReActor in Stable Diffusion to swap the original face onto the generated body. The tricky part is getting the alignment right. If the reference photo has Diana facing slightly left and your generated image has the subject facing straight ahead, the swap will look distorted. I had to regenerate about twelve base images before one came close enough in angle to make the swap look natural.
Another issue that caught me off guard: hair. Diana's hair was a significant part of her recognizable appearance, and AI generators tend to default to generic blonde hairstyles when they don't have specific instructions. I had to add detailed prompts about her signature layered blonde bob with the specific volume and parting style she had in the early 90s. Even then, the results varied. I ended up keeping the original hair texture from the reference photo and compositing it onto the generated image in Photoshop. That took about twenty minutes per image. The prompt engineering for this is more specific than most people realize. "Beautiful older woman" gives you a generic attractive senior. You need descriptors that constrain the output: "Caucasian female, approximately 60 years old, fair skin with natural aging, light blue eyes, refined facial features, contemporary casual clothing, soft natural daylight, photograph style, no studio lighting." The word "refined" is important here because it prevents the model from generating exaggerated or caricatured features. Diana had a distinctly structured face, and without that cue, the output drifts toward a generic elderly woman template. There's also a legal and ethical consideration worth noting. The royal family has not officially approved or endorsed any commercial use of Diana's likeness through AI generation. If you're doing this for personal or editorial purposes, you're generally fine. If you plan to sell prints or use it commercially, that's a different question. I've seen several creators get takedowns on platforms for this reason. Stick to non-commercial use unless you've consulted someone who understands the legal landscape.
Get the Full Details
The whole process — from reference image to final composite — took me roughly four to six hours for a single high-quality result. Most online tutorials claim you can do it in thirty minutes, and they're either using a pre-made template or they're dramatically oversimplifying the quality control that goes into a believable result. The difference between a decent output and a good one is mostly in the iteration count and the compositing step. If you're new to this, start with Midjourney's image prompt feature. Upload your reference photo directly into the chat and let it generate variations before you start swapping. This gives you a baseline to understand what the model does with that face structure. Then move to face-swapping tools once you have a generation you're happy with. Don't reverse that order — swapping onto a bad base just amplifies the problems. I also recommend keeping a spreadsheet of your prompt variations. After ten or fifteen attempts, you'll notice patterns in what works. For instance, adding "editorial photography style" to the prompt consistently produced better results than "photorealistic," which I found counterintuitive at first. The model seems to interpret "photorealistic" as a general aesthetic cue, while "editorial photography style" constrains it to a more specific lighting and composition framework. That alone cut my iteration time in half on later projects.
The results will never be perfect. No amount of prompt tuning will make an AI-generated aged photo indistinguishable from a real photograph of Diana at sixty. The technology simply isn't there yet for that level of fidelity, especially with a person whose face the public knows so intimately from decades of media coverage. But for editorial use, documentary research, or personal exploration, it's a workable process if you're willing to put in the iteration time.
