Yes, with one hard limit: the picture must show what the customer will actually receive. Delivery apps ask for photos that reflect the item served, and advertising rules treat a photo as a claim about size, quantity and ingredients. AI is low-risk for lighting, backgrounds and cropping a real photo, and high-risk for generating the dish itself.
Most trouble comes from small drifts rather than outright fakes: an extra patty, nine wings in a six-wing portion, a garnish the kitchen doesn't use. A label saying "AI-generated" doesn't repair any of those, so the useful question for a restaurant is what an AI picture is allowed to change, and what it must leave exactly as the plate leaves the pass.
What the apps, Google and the advertising rules say
Start with the app that matters most to you. DoorDash's merchant photo guidance, as of October 2026, asks for landscape 16:9 item photos of at least 1400 x 800 pixels, with the dish centred, at least 80% of it visible and filling roughly half to 70% of the frame. Its list of rejection reasons also covers vertical or square photos, people in shot, text overlays, borders and watermarks, stock or copied images, and creative or photoshopped backgrounds. The sentence that matters most for AI is this one: images "must reflect the actual item served, not an idealized or generic version".
DoorDash's photo pages don't mention AI-generated images by name, but the same rejection list says to "use real photos of your actual item" and rules out clip art, illustrations and stock graphics, which leaves little room for a dish generated from a description. The pages also disagree on the maximum file size (2 MB on the rejection page, higher figures elsewhere), so treat the upload screen as the final word and read the current rejection reasons in your merchant account before you upload. DoorDash's own AI tool is telling. Its help article pitches AI Photo Enhance as cleaning up backgrounds, improving lighting and sharpening images "all while preserving the appearance of your food", and enhanced photos still go through the normal photo review. One of its three modes, AI Replate, re-stages the dish on premium dinnerware, which deserves a second look if your food actually leaves in a clamshell. The platform draws its own line where I'd draw yours: AI around the food, not on it.
Google goes further and names AI. Its photo tips for Business Profiles ask for photos with "no significant alterations or excessive use of filters or AI. In other words, the image should represent reality." An invented dining room, or a dish you don't serve, fails that sentence on its face. The same page asks food businesses for at least three photos of the food or drinks they serve, so fill a thin profile with phone photos rather than generated ones. Other delivery apps set their own ratios and file sizes, so don't reuse one app's crop everywhere, and a website or printed menu isn't bound by any app's rules at all. It is bound by advertising rules.
Those rules haven't been rewritten for AI, and they apply whether or not AI was used. One advertising code puts the label question plainly: qualifications "may clarify but must not mislead by contradicting the claims that they qualify". A food picture is a claim in its own right about size, quantity, ingredients and appearance, even when the menu wording is careful, so a line of small print can't undo a picture of a bigger portion. The working rule is short: sell the dish you show, and show the dish you sell.
If you have customers in the EU, the AI Act adds a separate duty in Article 50(4): deep fakes, meaning AI-made or AI-altered images that look like real objects, places or events and could pass as genuine, must be disclosed. A qualified adviser can tell you whether a menu photo of your own dish falls inside it. The cautious course is to label AI artwork and avoid generated pictures of the dishes you sell.
| Where the picture appears | What the rule asks | Safe use of AI |
|---|---|---|
| DoorDash menu photo | Real photos of the actual item served; landscape 16:9; no illustrations, overlays, watermarks or creative backgrounds | Lighting and plain-background clean-up of your own photo, checked against the real dish |
| Other delivery apps | Their own ratios, sizes and rejection reasons; read each app's photo page | The same approach as above, once you have checked that app's rules |
| Google Business Profile | No significant alterations or excessive use of filters or AI; the image should represent reality | Light correction only; no generated interiors or dishes |
| Website and printed menu | General advertising rules: a picture is a claim | Real or lightly enhanced photos for dishes; labelled AI artwork for mood only |
There is more on writing the matching descriptions and handling photos in the tutorial on rewriting takeaway delivery app menus.
Ten details an AI dish photo tends to get wrong
Generators fill gaps with a generic version of the dish. The table lists the ten gaps I'd check on every picture, with what to compare each one against.
| Detail | How an AI picture drifts | Compare it with |
|---|---|---|
| Portion size | Fills the plate edge to edge, or makes the dish look larger against a hand or fork | The weight on the recipe card and the real plate diameter |
| Piece count | Nine wings, ten dumplings or eight sushi pieces where the portion has six | The count on the menu line |
| Layers and fillings | Extra patties, extra cheese, a second layer of toppings | The spec card, item by item |
| Garnish and props | Herbs, lime wedges, edible flowers, skewers and flags the kitchen never uses | Your plating guide |
| Sides and drinks | Fries, dips or a drink shown beside a dish that includes none of them | The item's actual inclusions on the menu |
| Cooking state | Perfect char marks, uniform golden crust, a pink centre that isn't on your menu | A photo of the dish straight from the pass |
| Sauce and shine | Dripping sauce, glossy oil, steam and smoke added for drama | How the dish looks 10 minutes after plating |
| Container | A ceramic plate or wooden board where the order travels in a foil tray or clamshell | What the delivery bag actually holds |
| Colour | Herbs greener and tomatoes redder than the real ones | The original photo, side by side |
| Allergen cues | Shows or hides sesame, nuts, cheese or egg in ways your recipe doesn't | Your allergen sheet |
The last row deserves a second look. Allergen information belongs in words, and the picture must never contradict it. A bun that shows sesame seeds when yours has none is a nuisance. A dessert that looks nut-free when it contains nuts is a safety problem for someone who relies on what they see. For allergen questions asked in words, see the tutorial on AI chatbots and allergen questions.
A 36-item menu, sorted by what AI may touch
Take a 40-seat bistro with 36 items on its delivery listing and no photographer in the budget. The owner's first thought is to generate all 36 pictures from text prompts. The sorting step below changes the plan. Each item goes into one of three groups before anyone opens an image tool. The numbers are illustrative.
| Group | Items | Examples | What AI may do |
|---|---|---|---|
| Clean-up only | 21 | Burgers, wraps, bowls, pizzas, single plated mains | Replace the background, even out the light, crop to 16:9; nothing on or around the food |
| Real photo, no AI | 9 | Wings, tacos, sharing platters, anything sold by the piece | Nothing; count and layers matter too much |
| Words only for now | 6 | Weekly specials and seasonal dishes | Nothing; photograph the first real batch, then publish |
Now the time. Generating all 36 from prompts takes about eight minutes an item to get a usable image after a few attempts, plus four minutes to check each against the spec card: 36 x 12 = 432 minutes, about seven hours. The sorted plan uses a phone for the 30 items that get photos. Plating and shooting take about six minutes each, so 180 minutes. AI clean-up on the 21 items takes about three minutes each, so 63 minutes. Checking all 30 against the spec card takes two minutes each, so 60. That is 303 minutes, about five hours, with no generated food anywhere on the page and six weekly specials that never needed a photo.
So the sorted plan is about two hours quicker once checking is counted, and it leaves the owner with a menu where every image has been compared with the plate. Generating everything only looks faster until the checking is added. The editing tool costs little. Photoroom's Pro plan, for example, lists at $12.99 a month ($7.50 a month billed yearly) for 1,000 exports and includes AI backgrounds, which is plenty for 21 clean-ups. The wider cost question, photographer or AI, is worked through in the tutorial on AI product photos versus a photographer.
One dish's spec card, and the audit of a generated picture
A spec card is a short description of the dish as it leaves your kitchen. This one is filled in for a burger (illustrative):
DISH SPEC CARD - Smoky chicken burger (menu item 14)
Served as: cardboard clamshell, paper wrap, no side
Bun: plain brioche, about 11 cm across, toasted, no seeds
Protein: 1 fried chicken thigh fillet, about 140 g cooked
Cheese: 1 slice cheddar, melted
Toppings: 3 pickle slices, 20 g slaw. Nothing else.
Sauce: 1 tablespoon house sauce, under the bun, never on top
Garnish and props: none (no skewer, no flag, no herbs)
Sides: none included (fries are a separate add-on)
Allergens (from the recipe sheet): gluten, milk, egg. No sesame.
Approved photo: burger-14-2026-10-27.jpg, signed off by [name]Now suppose someone generates a picture from the prompt "smoky chicken burger, professional food photography". Here is the audit against the card (illustrative):
| Check | Spec card says | Generated picture shows | Result |
|---|---|---|---|
| Protein | 1 chicken fillet | Two thick patties | Fail |
| Cheese | 1 slice, melted | 1 slice, melted | Pass |
| Toppings | 3 pickles and slaw | Bacon, tomato and lettuce as well | Fail |
| Bun | Plain, no seeds | Glossy, with sesame seeds | Fail |
| Sauce | Under the bun | Dripping down the side | Fail |
| Props | None | Cocktail stick with a flag | Fail |
| Side | None included | Basket of fries beside it | Fail |
| Container | Cardboard clamshell | Wooden serving board | Fail |
| Scale | Bun about 11 cm, hand-sized | Looks about 16 cm across | Fail |
One pass in nine isn't surprising for a generator that only knew the words "chicken burger". It filled the gaps with a generic burger, which is precisely what an "idealized or generic version" rule is aimed at. The sesame seeds are the worst failure: they contradict the allergen line on the card. A real photo of the clamshell, cleaned up by AI, would pass every row, because the only things the AI was allowed to change (background, light and crop) aren't rows on the card.
An edit prompt that leaves the food alone
When you do use an image tool on a real photo, the instruction has to forbid what these tools like to add. This is the wording I'd start from:
Edit the attached photo of our chicken burger. Keep the food, the packaging, the portion size and every ingredient exactly as photographed. Do not add, remove, move, resize or recolour anything on or around the food.
Change only these three things:
1. Replace the table surface with a plain, matt, mid-grey background.
2. Even out the lighting and soften harsh shadows, without changing the colour of the food.
3. Crop to 16:9 landscape with the burger centred and filling about 60% of the frame.
Do not add steam, sparkle, sauce drips, garnish, props, text, logos or people.A realistic result (illustrative): the tool returns a tidy grey background and softer light, but a lime wedge now sits on the clamshell lid, the chicken crust is a shade browner than the original and a few sesame seeds have appeared on the bun. The first is a prop, the second a colour change and the third an allergen cue, and all three would fail the audit. The fix is to reject that version and run it again with the unwanted items named ("no lime wedge, no seeds"), and if the tool keeps adding things, to stop using it for that dish and do the background with a conventional cut-out tool instead.
To check an edit, open the original and the edited image at the same zoom and flick between them. Count anything countable. Look at the edges of the food: edges that move, soften or sprout new detail are the giveaway. Free tools sometimes stamp a visible mark on downloads, and delivery apps reject watermarked photos, so check the file before you upload.
What a wrong photo costs, and how it shows up
Say the loaded-fries photo was generated from a description and shows about double the real topping. Three customers a week message to say the fries arrived plainer than pictured, and each gets a refund averaging $11. That is $33 a week, about $1,700 a year, before you count any one- or two-star ratings that mention the picture. The figures are made up, but the shape is realistic.
It rarely announces itself. The same dish keeps turning up in complaint messages and in reviews ("not like the picture") for weeks before anyone connects it to the photo. Reading your reviews in bulk, rather than one at a time, will surface it sooner. When it does arrive as a message, a reply like this keeps it short (illustrative):
Hi [first name], thanks for letting us know, and sorry the fries didn't match the photo. I've refunded [amount] to your original payment method. We're re-checking our menu pictures this week so they show exactly what we send out. [your name], [restaurant name]Then fix the photo, not only the complaint. The tutorial on answering restaurant complaints with AI covers the wider routine.
When a label helps and when it doesn't
A label explains how an image was made. It doesn't correct a false impression of what is being sold, and the advertising-code rule quoted earlier says as much. So labels belong on pictures that don't make a claim about a dish, and they never rescue a picture that does. Three wordings that are honest, and one that isn't:
- Website mood image: "Illustration created with AI. Our dishes are photographed as served."
- Real dish, lighting and background adjusted: "Photos show our actual dishes, with lighting and background adjusted."
- Social post using generated artwork: "AI-generated artwork" in the caption, and kept away from posts about a specific dish.
- The one that isn't: "Serving suggestion" under a picture that shows more than the portion contains.
Concept images for a dish you haven't launched are the awkward case. A labelled teaser is defensible, but it shouldn't sit on a menu or delivery listing, and never next to a price. The tutorial on a café test of AI-generated marketing photos looks at which pictures read as honest. For Google, keep AI out of interiors and dishes entirely, which also fits the advice in the tutorial on AI posts and photos for a restaurant's Google profile.
The ten-minute real photo
If a dish fails an AI treatment, or you simply don't trust one, the fallback is a phone photo taken properly. It takes about ten minutes for one dish once you have done it twice.
- Set up beside a window, with no flash and no overhead kitchen lights, on a plain table.
- Plate the dish exactly as the spec card says, in the packaging it leaves in, and weigh it once to confirm the portion.
- Wipe the rim and remove crumbs. This is the only styling you are allowed, because it is what a server would do.
- Shoot in landscape from about 45 degrees above, with the dish centred and filling half to 70% of the frame.
- Take three shots and keep the sharpest. Name the file with the item and the date.
- Save the spec card beside the photo, and note who signed it off.
That matches DoorDash's own framing numbers, so the photo is ready for the app without cropping tricks. If the real food looks plain, change the plating, the plate or the light before you change the picture.
Five questions for a supplier selling AI food photos
Agencies and freelancers increasingly offer "AI food photography" at a low price per image. If you outsource, ask these before you pay, and treat a dodged answer as an answer.
- Does every image start from a photo of my real dish, or is it generated from a description?
- Will you send the original beside the edited version, so I can compare them?
- Which tool do you use, and does it leave a watermark on the file?
- Who checks portion, ingredients and packaging against my spec card, and is that checking included in the price?
- Will you redo anything I flag at no charge, and will you delete my originals and edits if I leave?
A good supplier will welcome the spec card, because it tells them what not to improve. One who says the pictures "look better than the real thing" has just described the problem this tutorial is about.
Keeping photos true as the menu changes
A photo is accurate on the day it is approved and stale when anything behind it changes. Five triggers mean a re-shoot: a recipe change, a new supplier whose product looks different, new packaging, a portion change, or a change in what comes with the dish, such as a side added or dropped. Name one person who signs off every photo, so that approval doesn't depend on whoever uploaded it.
Once a quarter, spend 15 minutes on an audit. Pick five menu items at random, take the dish off the pass and hold it next to the listing. If a regular customer would notice the difference, the picture is wrong. Keep the original, the edited file and the spec card for each one, so if an app or a customer queries a picture you can show what you started with and what changed.
The rule that sits under all of this is short. Use AI around the food, never on it, and test every picture against the plate. If you can't, the real photo is the answer.
Questions restaurant owners ask about AI menu photos
Does DoorDash allow AI-generated menu photos?
DoorDash's photo pages don't name AI-generated images, but they ask for real photos of your actual item, showing the item served rather than an idealised or generic version, and they reject illustrations, stock graphics, creative backgrounds, overlays and watermarks. A photo generated from a description is the likeliest to fail that test, while a real photo with cleaned-up lighting is the safer route. Check the current rejection reasons in your merchant account before uploading.
Do I have to label AI food photos?
A label isn't the real test. Advertising rules look at whether the picture misleads, and one advertising code says a qualification may clarify a claim but must not contradict it. Label AI artwork used for mood, avoid generated pictures of your own dishes, and if you have customers in the EU, ask an adviser whether the AI Act's deep-fake disclosure duty reaches your images.
Can I use AI to remove crumbs or a stray fork from a real photo?
Usually yes, because you would wipe the same things off the plate before serving. The line is whether the edit changes what the customer receives. Removing a fingerprint on the rim is fine; removing a sauce spill the kitchen really sends out, or adding a garnish, is not. Keep the original file and compare the two before you publish.
What if my real food photographs badly?
Fix the conditions before the food: window light, a plain plate, a 45-degree angle and plating exactly as the spec card says. If the dish still looks flat, change the plating or the recipe, not the picture. An honest, slightly plain photo avoids the 'not like the picture' complaints that a glossy mismatch invites.
Can I show a dish that hasn't launched yet with an AI image?
For a teaser, a clearly labelled concept image is defensible, as long as it doesn't sit on the menu or delivery listing as if it were the dish. Photograph the first real batch before the item goes on sale and swap the picture then. Never put a generated image next to a price.
Further reads
- Is It OK to Edit Hotel Photos With AI? Fair Edits vs Misleading — The same fair-edit versus misleading-edit line, drawn for hotel photos.
- AI Tools for Ecommerce Photography: Backgrounds, Shadows, Consistent Sets — Background and shadow tools in detail, for any product on a plain surface.
- How to Write Menu Descriptions With AI That Still Sound Like You — Menu wording that matches the photo and still sounds like you.
- How to Create Product Photos With AI on a Small Budget — A budget photo workflow that also suits packaged food.
- Best AI for the Food Industry: Picks for a Cafe, Caterer or Takeaway — Which AI tools suit a café, caterer or takeaway, by job.
- What Are Diners Really Saying? Using AI to Analyse Your Reviews — Spot 'not like the photo' complaints by reading reviews in bulk.
- Delivery App Commission: Should Takeaway Menu Prices Rise on Apps? — A five-dish margin model for deciding whether to mark up delivery-app menus, with the order loss you can absorb and the price-parity rules to check.
- AI Restaurant Menu Maker: Printable and QR Menus in Canva — Build a printable and QR menu in Canva with AI help, then run a price and allergen check before the print run.
- AI for Photographers: Same-Day Quotes and Shot Lists From a Brief — Brief in, quote out the same day: a worked furniture-workshop shoot with an extraction prompt, a timed shot list, a priced quote and the pre-send checks.
- AI for a Cloud Kitchen: Keep Every Brand's Menu Consistent on Apps — A brand-by-app checklist for cloud kitchens: one master menu sheet, AI rewrites that keep the facts, sheet checks for prices and allergens, and a change log.
- Food Truck Vendor Applications: Draft Festival Forms With AI — A drafting method for festival and market forms: facts sheet, requirements table, answers from your facts only, and a gap check on insurance and licences.
- Free AI Restaurant Menu Generator: What the Free Tiers Allow — Canva Free, ChatGPT and Gemini compared for menu making: AI allowances, watermarks, print resolution and the free route that works.
- Can AI Handle Restaurant Menu Translation Without Mistakes? — How to translate a restaurant menu with AI: keep dish names, back-translate every line and have allergen wording approved by a person before it is printed.
- What Is a Ghost Kitchen? How an Existing Restaurant Can Test One — What a ghost kitchen is, and how a restaurant can test one from its current kitchen on one delivery app: menu, per-order economics and a four-week log.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: DoorDash merchant photo guidance (photo rejection reasons and photo requirements, October 2026) and its help article on AI Photo Enhance; Google Business Profile photo guidance and photo tips; Photoroom pricing page; Article 50 of the EU AI Act; an advertising code's rule on qualifications.