AI Food Photography: Generate Restaurant-Quality Dish Shots Without a Camera

Food photography has always been one of the most expensive line items in a restaurant’s marketing budget. You’re paying a photographer, a food stylist, a prop stylist, and a studio rental — just to get twelve shots of a pasta dish that’ll live on Instagram for three days. The math has never made sense, and AI image generators are finally offering a credible alternative.

This isn’t a pitch for cutting corners. Done wrong, AI food images look like they were rendered inside a video game from 2019 — plastic textures, uncanny steam, garnishes that defy physics. Done right, they’re indistinguishable from a professional shoot. The difference is in how you prompt, which tool you reach for, and what parameters you dial in. This tutorial covers all of that, from the first prompt to the final export.

We’re using Midjourney V7 as the primary workhorse here (the photorealism improvements in V7 are significant enough to make V6 feel like a demo), with Firefly and Leonardo as legitimate alternatives depending on your specific use case. Every prompt below is copy-paste ready.

What You’ll Actually Achieve

By the end of this tutorial, you’ll be able to generate plated dish photography with controlled lighting direction, believable depth of field, realistic surface textures on food, and composable layouts you can iterate on without reshooting. You’ll also understand why most AI food photos fail — which is more useful than any single prompt.

What You’ll Need

A Midjourney subscription (Basic plan works, Standard gives you more fast hours), optionally a Firefly subscription or Leonardo AI account for texture and style control, and a basic understanding of food photography vocabulary like catch light, hero angle, and fill ratio. You don’t need to be a photographer, but knowing what a 45-degree overhead shot means will help you speak the right language to these models.

Step 1 — Nail Your Hero Angle First

Before you write a single prompt, decide on your camera angle. Food photography has three standard working positions, and each reads differently. Overhead (flat lay, 90 degrees) works for grain bowls, pizzas, and anything with surface detail you want to show. The 45-degree angle — sometimes called the three-quarter view — is the workhorse of restaurant menus, showing height and texture simultaneously. Straight-on (eye level) suits drinks, burgers, and anything with dramatic vertical stacking.

Pick your angle before you prompt, because adding it explicitly makes the model’s job much easier and your results dramatically more consistent.

Step 2 — The Base Prompt Structure That Actually Works

Forget poetic food descriptions. Models respond to technical photography language, not adjectives about how delicious something smells. The structure that produces the most consistent restaurant-quality results follows this pattern: [dish name] + [plating style] + [camera angle] + [lighting setup] + [surface/background] + [lens/depth of field details] + [mood/color grade] + [technical quality flags].

Here’s the foundational prompt for a pasta dish shot:

close-up food photography of handmade tagliatelle with truffle cream sauce and shaved black truffle, plated in a shallow white ceramic bowl, shot from 45-degree angle, single soft-box lighting from upper left creating gentle shadow on right side, dark walnut wood table surface, 85mm lens, f/2.8 shallow depth of field with focus on front pasta strands, warm color grade with slightly desaturated shadows, editorial food photography style --ar 4:5 --style raw --v 7

The --style raw flag is doing important work here. Midjourney’s default stylization tends to over-saturate food and add a slight illustration quality. Raw mode keeps the output closer to actual photography. The --ar 4:5 ratio is Instagram’s native portrait format — if you’re shooting for print menus, switch to --ar 3:4 or --ar 1:1.

Pro tip ✅

Always specify the light source direction explicitly — “from upper left,” “from camera right,” “backlit from behind.” Without this, models default to flat, even lighting that looks like a food delivery app thumbnail. A single defined light source creates the shadows that make food look three-dimensional and real.

Step 3 — Control Your Lighting Like a Pro

Lighting is where most AI food photography falls apart. The model defaults to something that technically illuminates the food but produces no drama, no texture, and no appetite appeal. Here are four proven lighting setups to embed directly in your prompts:

Soft window light (bistro feel):

overhead flat lay food photography of acai bowl with granola, fresh berries, and honey drizzle, shot directly above, diffused natural window light from the left side casting soft shadows, light gray linen surface, 50mm equivalent, very shallow depth of field with center focus, muted pastel color palette, Scandinavian minimalist food styling --ar 1:1 --style raw --v 7

Dramatic single-source (fine dining):

food photography of seared duck breast with cherry jus and microgreens, plated on black matte ceramic, shot from 35-degree angle, single hard spot light from upper right creating strong directional shadows, deep black background, 100mm macro lens, f/4 with rear elements falling out of focus, high contrast dark-mood editorial style, Michelin star restaurant plating --ar 4:5 --style raw --v 7

Backlit golden hour (social media optimized):

food photography of crispy fried chicken sandwich with pickles and sauce drip, shot at eye level from straight front, warm backlight creating rim light and translucent effect on lettuce leaves, rustic wooden surface, condensation visible on accompanying drink glass, 85mm portrait lens, f/2.0 bokeh background, golden warm tones, appetite-forward fast casual style --ar 4:5 --style raw --v 7

Three-point fill (bright and airy brunch):

bright airy food photography of eggs benedict with hollandaise sauce on toasted English muffin, steam rising, 45-degree angle shot, soft three-point studio lighting with white fill card reflecting on left, white marble surface with fresh herbs scattered, 85mm lens f/3.5, light and fresh color grade, lifestyle brunch photography style --ar 4:5 --style raw --v 7

Pro tip ✅

Mentioning a specific surface material is one of the highest-leverage additions to any food prompt. “Dark walnut wood,” “honed white Carrara marble,” “slate tile,” “light oak,” and “weathered concrete” each produce completely different moods and interact with the light source differently. Surface texture also gives the model something to render shadows onto, which makes the composition feel grounded rather than floating.

Step 4 — Lock Composition, Then Iterate the Garnish

One of the most useful workflows for restaurant clients is generating a composition you like, then varying just the garnish or plating detail without changing the framing. In Midjourney V7, you do this with the Vary (Subtle) option after upscaling, combined with a reference image seed.

First, generate your base composition and note the seed number (click the envelope emoji on Discord or use --sref with the image URL in the web app). Then iterate with slight prompt modifications:

food photography of grilled salmon fillet with lemon butter sauce, plated on white oval ceramic, shot from 45-degree angle, soft window light from upper left, dark slate surface, 85mm lens f/2.8, garnished with fresh dill sprigs and capers, editorial food photography --ar 4:5 --style raw --v 7 --seed 847392

Now change just the garnish element while keeping everything else identical:

food photography of grilled salmon fillet with lemon butter sauce, plated on white oval ceramic, shot from 45-degree angle, soft window light from upper left, dark slate surface, 85mm lens f/2.8, garnished with micro herbs and lemon zest strips, editorial food photography --ar 4:5 --style raw --v 7 --seed 847392

Using the same seed keeps the underlying composition locked while the model explores the garnish variation. You won’t get pixel-perfect consistency, but the camera angle, lighting direction, and plate position stay substantially similar — which is exactly what you want when presenting options to a client.

Warning ⚠️

Midjourney V7 still struggles with certain food physics — sauces sometimes look like plastic, carbonated drinks lose their bubbles, and anything involving visible steam can get weird. If a specific element keeps rendering incorrectly, try describing its texture explicitly: “sauce with visible sheen and slight transparency,” “carbonated water with fine bubbles visible in glass,” “light wisps of steam rising from surface.” Pretend you’re directing a photographer to notice something specific, not just describing what the food is.

Step 5 — Leonardo AI for Texture Control

When Midjourney isn’t giving you the surface texture you need — particularly for bread crust, meat sear marks, or melted cheese — Leonardo AI’s PhotoReal mode is worth a detour. Its texture rendering on organic materials often surpasses Midjourney’s defaults for close-up macro shots.

extreme close-up macro food photography of freshly baked sourdough bread loaf, cracked crust with flour dusting, golden brown with visible air pockets in cross-section, shot straight on, warm natural side lighting revealing surface texture, dark background, 100mm macro lens, incredibly detailed crust texture, editorial artisan bakery photography

In Leonardo, set PhotoReal to On, Alchemy to On, and Photo Real Strength to 0.55–0.65 (higher values tend to over-crisp the output). For bread and pastry specifically, Leonardo consistently produces more believable crust detail than Midjourney’s defaults.

Note 💡

Adobe Firefly (currently Firefly Image 3) is the better choice when you need generated food images that you’re certain are commercially safe — Firefly trains exclusively on licensed Adobe Stock content. The tradeoff is that photorealism maxes out slightly below Midjourney V7 in most food categories. For small restaurant clients who need to be absolutely clean on IP, Firefly is the right call regardless of the slight quality gap.

Step 6 — The Depth of Field Formula

Shallow depth of field is what separates menu photography from food blog snapshots, and getting it right in prompts requires specificity. Don’t just write “blurry background” — models interpret this inconsistently. Instead, describe what’s in focus and what’s falling off:

food photography of beef tartare with quail egg yolk and capers, plated in a ring mold on black slate, shot from 45-degree angle at table level, single soft box from upper left, sharp focus on the quail egg yolk in foreground with tartare slightly soft behind it, background table elements entirely defocused into smooth bokeh circles, 85mm portrait lens f/1.8 rendering, editorial fine dining photography --ar 4:5 --style raw --v 7

Specifying the focal point as a specific element (“sharp focus on the quail egg yolk”) gives the model a target. Specifying background behavior separately (“smooth bokeh circles”) tells it how to treat the falloff. This two-part depth of field instruction produces dramatically more controlled results than generic blur requests.

Pro tip ✅

For drinks and cocktails specifically, always mention condensation if the glass should be cold, and specify ice cube clarity. “Crystal clear ice cubes with slight reflections” vs. “cloudy ice with visible cracks” produces completely different drinks — one looks like a premium cocktail bar, the other looks like a dive. The model knows the difference if you tell it which you want.

Step 7 — Export and Delivery Workflow

Midjourney’s upscale output at V7 runs around 1456×1456 pixels at baseline, with Upscale (2x) taking you to approximately 2912×2912. For most Instagram and social use this is sufficient, but for print menus you’ll want to run the output through an upscaler. Topaz Gigapixel AI and Adobe’s Super Resolution (built into Lightroom and Camera Raw) both handle AI-generated food images well — Gigapixel’s “Standard” mode works best on these outputs; the “Art & CG” mode tends to over-sharpen the organic food textures.

For color correction, import into Lightroom and apply a conservative edit: pull back the vibrance slightly (food AI tends to over-saturate reds and oranges), bump up the whites to add air, and apply a slight curves S-shape to add separation in the midtones. The goal is making it look like a well-exposed photograph, not like a render — which means if it looks too perfect, dial it back.

Export as JPEG at 85% quality for digital use, and as TIFF at full resolution if the client needs it for print. At 300 DPI, a 2912-pixel image gives you roughly 9.7 inches — sufficient for a standard menu card or a half-page ad.

Avoid 🚫

Don’t use AI-generated food images for anything claiming to depict a specific real dish without a disclaimer, especially for allergen-sensitive contexts. The model might add an ingredient that isn’t actually in the dish, and the image won’t match what comes out of the kitchen. For regulatory and customer trust reasons, use AI food photography for aspirational brand imagery and menu design — not as a legal substitute for accurate dish representation in markets where that matters.

The Prompt Cheat Sheet — Quick Reference

Five prompts you can adapt immediately for the most common restaurant use cases:

Burger (fast casual):

food photography of double smash burger with melted American cheese, caramelized onions, and special sauce, shot at eye level from straight front, warm backlight creating golden rim light on sesame seed bun, condensation on accompanying cola glass, rustic dark wood surface, 85mm lens f/2.0, appetite-forward casual dining style --ar 4:5 --style raw --v 7

Cocktail (bar menu):

drinks photography of classic negroni cocktail in crystal rocks glass, large crystal clear ice cube, orange peel garnish, shot at slight downward angle, moody bar ambient lighting with single candle catch light visible in glass, dark marble bar surface, 85mm lens f/2.8 shallow focus on orange peel, warm dark tones, upscale cocktail bar aesthetic --ar 4:5 --style raw --v 7

Dessert (pastry close-up):

macro food photography of chocolate lava cake with vanilla ice cream melting and chocolate sauce dripping, cross-section cut revealing molten center, shot from 35-degree angle, warm soft window light from left, white ceramic surface, 100mm macro lens f/4 with ice cream in sharp focus and background elements soft, warm rich color palette, patisserie editorial photography --ar 4:5 --style raw --v 7

Why This Actually Works for Restaurants Now

The quality bar crossed a threshold in 2025 where the gap between a well-prompted AI food image and a mid-range food photography shoot became difficult to distinguish at Instagram viewing distances. That’s not an argument for never hiring a photographer — it’s an argument for being strategic. Use AI generation for the thirty menu items you change seasonally, for social content that lives for 48 hours, and for testing how a new dish presents before spending studio time on it. Use real photographers for your hero shots, your website banner images, and anything that will get scrutinized at large scale.

The restaurants getting the most out of this aren’t replacing their studios — they’re using AI generation to do the work that wasn’t getting done at all because the studio budget ran out. That’s a genuinely useful thing, and the prompts above will get you there without needing a single softbox.

Promptyze

Promptyze covers generative AI in plain English — hands-on reviews, tutorials and daily news, fact-checked and hype-free.

See Full Bio