A Realistic Photo Experiment.
I tested an experimental AI image generation prompt designed to preserve facial identity from a reference photo while changing the scene, pose, clothing control, lighting, and photographic style.
What Was I Actually Testing?
This experiment started with a simple question: can a detailed AI image generation prompt preserve the recognizable identity of people from a reference photograph while creating a completely different photographic scene?
Instead of using a short instruction such as “make a realistic couple photo”, I built the prompt in several layers. Identity comes first, followed by pose and interaction, getup and dress, environment, lighting, camera characteristics, texture and finally instructions designed to reduce common image-generation errors.
The goal was not to create a magic formula. I wanted to see how much control could be achieved by describing the important visual elements more precisely.
Step 1 — The Reference Image
The original photograph is used as the primary visual reference. The prompt specifically asks the image model to retain recognizable facial characteristics instead of treating the photograph only as general inspiration.
Step 2 — Full AI Image Generation Prompt
Here is the complete experimental prompt. It is intentionally displayed inside a scrollable prompt viewer so the entire instruction can be read and copied without making the page extremely long.
Use the uploaded photo as the primary identity reference for both people. Preserve both faces as closely and faithfully as possible to the reference, including face shape, eyes, eyebrows, nose, lips, jawline, skin tone, natural asymmetry, age appearance, facial hair and distinctive features. Keep their expressions natural, candid and emotionally believable. Do not redesign, beautify, merge or replace their identities. Create an ultra-photorealistic lifestyle editorial portrait in a sunlit Kerala courtyard. The man stands slightly behind and to the right of the woman, gently leaning toward her and placing a soft affectionate kiss directly on the upper forehead, centered above her eyebrows and clearly away from her eyes, eyelids, nose, cheek and lips. His lips must visibly contact only her forehead. She tilts her head naturally toward him with softly closed eyes, chin slightly lifted and a genuine warm, relaxed smile. His left hand rests modestly at her waist while her arms remain relaxed at her sides. Preserve the couple’s original clothing exactly as shown in the reference photo, including the man’s maroon kurta and the woman’s original maroon saree, blouse, gold border, jewelry, bangles and hairstyle. Do not replace, recolor or redesign their outfits. Use a cream plaster wall with a dark wrought-iron decorative panel and graphic shadow on the left, dense tropical palm foliage on the right, and natural ground with a few dry leaves. Warm late-morning sunlight from upper left, realistic rim light, natural shadows and subtle jewelry highlights. Eye-level camera, vertical 3:4 composition, three-quarter body from mid-thigh upward, centered couple. Simulate 50mm at f/2.2, ISO 100, handheld candid photography with both faces sharply focused and the background softly separated. Use realistic skin pores, fine hair strands, fabric weave and environmental textures, natural warm skin tones and realistic dynamic range. High-resolution Full HD/4K, professional photorealistic quality. Avoid identity drift, facial distortion, plastic skin, excessive retouching, unnatural anatomy, malformed hands or fingers, artificial lighting, HDR, oversaturation, CGI or illustration.
Pose & Interaction — What Can Be Changed?
The pose is mainly controlled by the interaction and positioning instructions in the prompt. If you want to use the same AI photo editing prompt for a different pose, this is one of the first sections to modify.
- Position: Who stands in front, behind, left or right.
- Body posture: Standing, sitting, walking, leaning or turning.
- Body angle: Front-facing, side-facing or three-quarter position.
- Head direction: Looking at the camera, looking at each other or looking away.
- Head tilt: Natural tilt toward another person or a specific direction.
- Hand placement: Hands at the waist, shoulders, sides or holding another object.
- Interaction: Standing together, holding hands, hugging or another natural interaction.
- Expression: Smile, relaxed expression, serious expression or candid emotion.
- Camera relationship: How the subjects face and relate to the camera.
Getup / Dress — What Can Be Changed?
The clothing instructions are separate from the pose instructions. This makes it easier to reuse the same reference image prompt while experimenting with different outfits or maintaining the original clothing.
- Main outfit: Saree, kurta, shirt, suit, dress or other clothing.
- Color: Change the primary or secondary clothing colors.
- Fabric: Cotton, silk, linen, wool or another material.
- Design: Plain, printed, embroidered or traditional styling.
- Jewelry: Necklace, earrings, bangles, rings or other accessories.
- Hairstyle: Keep the original hairstyle or specify a new one.
- Footwear: Traditional, casual or formal footwear.
- Overall look: Traditional, casual, festive, formal or editorial.
Step 3 — The Tested Result
This is where the experiment becomes more useful than simply publishing the prompt. The image below is the actual test result I am using to evaluate how well the instructions work together.
Breaking Down the AI Image Prompt
The prompt may look long when read as one block, but each part has a specific purpose. This layered structure is what makes the experiment interesting.
Facial characteristics are described first so identity remains a primary part of the instruction.
The position of both people, body direction, head movement and interaction are explicitly described.
Clothing, jewelry, hairstyle, and colors are connected to the original reference.
Specific background elements are used instead of leaving the environment completely open.
Focal length, aperture, ISO, viewpoint and composition help define the intended photographic character.
Skin, hair, fabric and environmental details are included to push the result toward a natural photographic appearance.
The final instructions identify common unwanted results such as identity drift, distorted anatomy and plastic-looking skin.
The individual instructions work together to aim for the appearance of a naturally photographed editorial portrait.
What I Observed From The Test
The useful part of an AI image experiment is not claiming that a prompt is perfect. It is looking at which instructions appear useful and where the image model still has room to interpret the request differently.
Identity preservation is introduced before the environment, giving facial consistency a prominent position in the prompt.
Specific spatial descriptions provide more information about the intended pose than a simple request for a romantic couple photo.
Clothing instructions attempt to maintain continuity with the reference rather than allowing the new scene to completely redefine the subjects.
Camera and lighting language is used alongside the visual description instead of relying only on words such as “ultra realistic”.
Negative instructions can help communicate unwanted visual outcomes, but they should not be treated as a guarantee against generation errors.
One Important Limitation
A detailed prompt does not automatically provide perfect control. Different image-generation systems can interpret the same written instructions differently, particularly when a reference image, multiple people, physical interaction, and detailed composition are involved.
Because of that, I would not describe this as a “100% identity preservation prompt”. It is better understood as an experimental instruction set designed to improve control over an AI-generated image.
0 Comments