How to Generate Realistic Photos
Choose a model for realistic portraits using a documented three-model example. Build a clear prompt and check skin, anatomy, light, and reference needs.
Create a new photo or preserve a real person?
For an original fictional person, start with a text prompt. To keep a person from an existing photo, use that photo as an attached reference with a model that supports image input. A realistic-looking new face does not demonstrate identity preservation.
This guide covers text-to-image photography. For reference-led portraits and background changes, follow the portrait guide linked below.
Write the visible photographic brief
- Subject and action: say who is in the frame, what they are doing, and which expression matters.
- Composition: specify headshot or waist-up framing, eye level or another viewpoint, and room around the subject. Set the aspect ratio in the generator too.
- Light: name a direction, softness, and source, such as overcast window light from camera left.
- Materials: request natural skin variation, fabric texture, and believable shadows. Add grain or blur only when the scene calls for them.
- Camera terms describe an intended look; they do not prove that a particular camera, lens, or physical exposure was used.
What our same-prompt portrait test actually showed
On September 3, 2026, Nano Banana Pro, GPT Image 2, and Midjourney V8.2 each received the same text-only portrait prompt at 3:4. We kept the first returned image. All three passed the visible checks for subject, clothing, gaze, and absence of text or borders.
Nano Banana Pro framed more of the apartment; GPT Image moved closer to the face; Midjourney gave the most polished editorial finish. These are observations from one output per model, not a universal ranking, a success rate, or a test of keeping someone’s identity.
Copy the exact English prompt from that test
The prompt below was used in the dated test. Set the ratio to 3:4 and start without a reference image to match its setup. A new run can differ; this is not a recipe for an identical face.
Create a vertical 3:4 photoreal editorial portrait of a fictional adult woman in her late twenties, seated beside a large north-facing window in a quiet apartment. She wears a plain charcoal knit top, looks directly into the camera with a calm, self-possessed expression, and has natural skin texture with fine facial detail. Use soft overcast daylight, gentle shadow falloff, a 50 mm documentary perspective, eye-level framing, restrained neutral colors, and subtle fine grain. Keep the setting believable and uncluttered. No visible words, logos, branded objects, celebrity likeness, fantasy elements, beauty-filter skin, or decorative border.Check the result before judging the style
- Does the image keep the requested subject, action, expression, clothing, and framing?
- Do the light direction, catchlights, cast shadows, and reflections agree?
- Are skin and fabric plausible without waxy smoothing or excessive sharpening?
- Inspect visible hands, eyes, teeth, glasses, and object edges for distortion.
- When comparing models, hold the prompt, ratio, references, and selection rule constant. Record model version and date; separate compliance from your preferred look.