There is a moment every AI image creator knows: you generate an image that is technically perfect but obviously synthetic. The lighting is flat, the skin is too smooth, the geometry is subtly wrong. Then someone shows you a prompt that produces an image indistinguishable from a photograph, and you realize the difference was never the model. It was the prompt.
Writing prompts for photorealistic images is a craft with its own grammar. This guide breaks down the anatomy of a realistic prompt, explains the advanced techniques that separate beginners from professionals, and shows you how to adapt your prompts to different model families. Every section includes examples you can use immediately.
Why Photorealism Became the New Standard
The demand for photorealistic AI images is not a niche obsession. It comes from practical needs: e-commerce needs believable product shots, filmmakers need reference frames for virtual production, marketers need visuals that fit into real-world brand campaigns, and social media audiences have learned to reward authenticity.
The result is that generic prompts no longer cut it. An image that looks like a painting has its place in art projects, but when the goal is realism, every word in your prompt has a job to do. The models have the capability; the prompt decides whether you unlock it.
The Anatomy of a Photorealistic Prompt
A realistic prompt is not a sentence. It is a structured specification with four layers: subject, setting, lighting, and camera. Each layer answers a question the model cannot guess.
Subject and Setting
Start with the subject and setting, but be specific in a photographic way. "Portrait of a 30-year-old woman with silver jewelry" is a start. "Portrait of a 30-year-old woman with intricate silver jewelry, standing in a narrow market alley, late afternoon, vendors blurred behind her" gives the model context to anchor the scene.
Include details that a photographer would notice: age range, clothing material, posture, expression, environment, and time of day. Every concrete detail reduces the space in which the model improvises, and improvisation is where realism breaks down.
Lighting Is Everything
Lighting is the single factor that separates a real photo from a synthetic image. The best prompts borrow the vocabulary of professional photography: Rembrandt lighting, butterfly lighting, split lighting, three-point lighting, soft key light, rim light, golden hour, overcast diffusion.
Describe the direction, quality, and color of the light. A soft key light from the left with a warm rim light on the right tells the model exactly how the shadows should fall. When you specify lighting, faces get dimension, skin gets texture, and the image stops looking flat.
Camera and Lens
To make an image look shot on a real camera, specify camera parameters: lens focal length, aperture, depth of field, film stock, and grain. Phrases like "shot on Kodak Portra 400" or "85mm lens, f/1.8, shallow depth of field" carry real meaning to modern models.
Film stock references are especially powerful because they encode a color palette and grain structure that audiences associate with photography. Even if you have never touched the film, naming it signals the model to replicate its look.
Negative Prompts and Weighting
The positive prompt tells the model what you want. The negative prompt tells it what you do not want, and it is essential for photorealism. Artifacts, warped hands, plastic skin, oversaturation, and cartoonish features are all things you can push against.
Build a reusable negative list for realistic work: "plastic skin, airbrushed, oversaturated, cartoon, illustration, 3d render, warped fingers, extra limbs, watermark, text." Keep it stable across a project so your results stay consistent.
Weighting is the next level. Modern tools let you emphasize or de-emphasize specific terms. When the model keeps giving your subject the wrong eye color, weight the correct description higher. When a background element keeps dominating, lower its weight or move it to the negative prompt. Weighting is how you negotiate with the model instead of starting over.
Keeping Faces and Characters Consistent
Photorealism is about more than one image; it is about series. A character that changes face between images breaks the illusion completely. The practical solution is multi-image fusion: feed the model several reference frames of the same person from different angles, expressions, and lighting conditions, and it builds a unified identity.
This is the same technique used for video production, and it works for stills too. Before you generate a series, build a small reference set: a front view, a three-quarter view, a profile, and a version under the lighting you plan to use. The more consistent the references, the more consistent the character.
Adapting Prompts to Different Model Families
Every model family has its own temperament. The same prompt produces different results on different engines, and learning the differences saves you hours of frustration.
Flux and Runway: Detail Fidelity
Models in the Flux and Runway families reward detailed, photographic vocabulary. They understand lens and film references well and respond strongly to lighting specifications. Use precise terminology, stack your details deliberately, and reserve some of your weight budget for texture words: pores, fabric weave, dust particles, subtle motion blur.
Sora and Kling: Narrative Realism
The Sora and Kling families think in scenes, not just stills. They respond to context: what happened before this moment, what the subject is doing, what mood the scene carries. For realistic video frames from these models, write your prompt as a brief camera direction: subject, action, environment, light, camera movement, and the emotional beat.
PixVerse and MiniMax: Cinematic Control
These families shine when you lean into cinematic language. Depth of field, anamorphic look, lens flares, and deliberate color grading all work well. They also give you fine-grained control over camera behavior, so specify the shot type: close-up, medium, wide, crane shot, tracking shot.
Structured Prompting for Batch Workflows
When you need many images, consistency matters more than perfection. Build a prompt template with fixed and variable parts. The fixed part locks the style: lighting setup, film stock, lens, grain. The variable part changes per image: subject, pose, setting, props.
This is how professionals generate a whole catalog of product shots or a full storyboard. The fixed template guarantees the series looks like it was shot in the same session, while the variables provide variety. Keep the template in a document you can version, and log which seeds produced the best results.
Ten Example Prompts You Can Steal
-
Portrait: "Candid portrait of a 40-year-old man with a weathered face, soft window light from the left, 85mm lens, f/2, shallow depth of field, shot on Kodak Portra 400, subtle film grain."
-
Product: "Minimal product shot of a ceramic coffee mug on a wooden table, morning sunlight raking across the surface, macro detail on the glaze, 50mm lens, f/5.6, clean neutral background."
-
Street: "Rainy street at dusk, neon signs reflecting on wet asphalt, a lone figure with an umbrella walking away, motion blur on traffic, cinematic color grading, 35mm lens."
-
Still life: "Overhead shot of a rustic breakfast table, fresh bread, jam jar, copper kettle, soft overcast window light, natural shadows, shot on film, muted colors."
-
Portrait, dramatic: "Studio portrait of a dancer in profile, split lighting, black background, textured fabric costume, rim light on the silhouette, medium format look, fine grain."
-
Landscape: "Coastal cliff at golden hour, waves crashing below, warm light on wet rock, distant lighthouse, atmospheric haze, wide-angle lens, rich color, natural detail."
-
Macro: "Extreme macro of a dragonfly wing, dew drops catching backlight, iridescent detail, shallow depth of field, pure black background, scientific photography style."
-
Interior: "Cozy reading corner in an old library, afternoon sun through a tall window, dust motes in the light beam, leather armchair, wooden shelves, 24mm lens, warm tones."
-
Animal: "Close-up of a ginger cat staring at the camera, soft studio light, whiskers in sharp detail, dark neutral background, 100mm macro lens, natural fur texture."
-
Urban architecture: "Brutalist concrete building facade, low-angle shot, harsh midday sun creating geometric shadows, minimalist composition, fine architectural detail, no people."
Use these as starting points, not as finished work. Adjust the lighting and lens to your subject, and keep your negative prompt applied.
Common Mistakes and Fixes
The most common mistake is stopping at the first generated image. Realism is a numbers game: generate several options, compare, and refine. The model is a collaborator, and the first draft is rarely the best one.
The second mistake is ignoring resolution and upscaling. A realistic prompt can be ruined by a soft upscale. Generate at the highest native resolution your model supports, and if you upscale, do it in steps with a good upscaler rather than in one aggressive jump.
The third mistake is neglecting the negative prompt. Beginners focus entirely on what they want, then wonder why artifacts appear. A strong negative prompt prevents whole categories of errors before they happen.
Frequently Asked Questions
How many images should I generate per concept? Plan three to five options per approved image. For hero images in client work, expect more iterations before you lock one.
Do I need to know photography to write good prompts? It helps enormously, but you can learn the vocabulary quickly. Lighting names, lens focal lengths, and film stocks are a small set of terms that unlock large improvements.
Why do my images look too smooth? Add texture vocabulary to your prompt and check your negative prompt for terms that force airbrushing. Sometimes the culprit is the upscaler, not the generation.
Can I use one prompt template for every model? No. Adapt your prompt to the model family. What works beautifully on one engine may be ignored or misinterpreted on another.
Final Thoughts
Photorealism is not magic; it is specification. The models have the ability, and your prompt decides whether they use it. Learn the vocabulary of light, lens, and film. Build a reusable negative prompt. Keep references for characters and series. And above all, iterate: generate, compare, refine, repeat.
The difference between a synthetic-looking image and a believable photograph is not talent. It is the discipline of describing exactly what a camera would see. Master that, and the models will follow.
Building a Workflow for a Consistent Series
A single realistic image is a win, but a series is where the craft shows. Whether you are creating a product catalog, a storyboard, or a character study, the workflow is the same.
First, fix the style template: lighting setup, film stock, lens, grain, and color palette. Write it once and reuse it. Second, fix the character or subject with a reference set, using the fusion technique described earlier. Third, generate the stills for every scene in the series before you refine any of them, so you can see the whole set and catch consistency problems early. Fourth, only then iterate on individual images, keeping the template and the references locked.
This is the same discipline used in photography: one session, one light, one language. The series reads as a body of work instead of ten unrelated images, and the audience can trust what they are looking at.
Troubleshooting Common Artifacts
Even with good prompts, artifacts appear. Learn the fixes for the most common ones.
Warped hands and fingers: reduce the complexity of the action in the prompt, generate the hand at a larger scale, or move the hand out of focus. Plastic skin: add texture vocabulary and strengthen the airbrushing ban in the negative prompt. Oversaturation: lower the color terms or specify a muted palette and a named film stock. Blurry faces: increase the weight of the face description and generate at higher resolution. Text and watermarks: keep them in the negative prompt for every realistic image.
Keep a personal list of fixes. After a few projects, your list is faster than any tutorial, because it is written in your own workflow and your own subjects.
More Frequently Asked Questions
How do I keep a consistent style across many images? Use one fixed template for the style variables and change only the subject variables. Store the template in a document you can version, and keep the negative prompt identical across the series.
Do negative prompts slow down generation? No, they are processed in the same pass as the positive prompt. A strong negative list prevents whole categories of errors and usually saves iterations, not costs them.
