FRAI menu generator — photo to digital menu
An AI menu generator turns a photo of your printed restaurant menu into a fully structured digital menu — categories, items, prices, descriptions, all in your language — without you typing a single thing. Hayde's AI menu generator uses Google's Gemini 2.5 Flash vision model and runs during the free signup wizard.
How it works
Take a photo
A clear shot of your printed menu — JPEG is fine. Multiple pages? Take one photo per page.
Upload during signup
In the wizard, choose "Paper Menu" as your delivery type and upload the photo.
AI extracts the menu
Gemini reads the photo, identifies categories and items, pulls out prices, writes descriptions, and auto-translates to your selected language.
Review and publish
The wizard shows you the extracted menu. Fix any misreads in seconds. Hit publish.
What the AI extracts
Categories
Appetizers, mains, desserts — and any custom categories your menu uses. Each gets a unique ID and a properly-cased label.
Items
Every dish on the menu becomes a separate item with a name, price, and description.
Prices
Numeric prices in your currency, parsed cleanly from the photo. Decimal separators handled per locale.
Descriptions
If your printed menu has descriptions, Gemini extracts them. If not, you can leave the field empty or add them later.
Auto-translation
The extracted menu is automatically translated into your selected language during the same pass. Hebrew, French, Spanish, Bulgarian, Russian, Greek — pick when you sign up.
Title case
Item names get normalized to title case so "FISH SOUP" doesn't end up shouting at customers in lowercase or all-caps.
What the AI won't do
Honest list of limitations — better to know up front than be surprised mid-launch.
- Read text that's blurry, sideways, or covered by glare. Take the photo straight-on with good light.
- Read handwritten menus reliably. Type out a handwritten menu first, then take a photo.
- Extract dish images. Photos of dishes are not extracted from the menu photo — you can upload them separately.
- Process PDFs directly. Take a screenshot of each PDF page first.
- Set allergens automatically. Allergen tagging is a manual step after the AI import.
FAQ
How does the AI menu generator work?
You upload a photo of your printed menu during the Hayde signup wizard. The photo is sent to Google's Gemini 2.5 Flash vision model, which reads the menu and returns a structured JSON with categories, items, prices, descriptions, and translations. Hayde loads that JSON into your menu, ready to review.
Is the AI feature included in the free trial?
Yes. The AI menu generator runs during the free signup — before you commit to a paid plan. Your menu is fully built by AI on the free tier.
Which file types are supported?
JPEG images of your printed menu. If you have a PDF, take a screenshot of each page first. PDF input is on the roadmap but not shipped yet.
How accurate is the extraction?
On a clear, well-lit photo of a typed menu, Gemini extracts categories and items with very high accuracy. Prices are extracted cleanly. Expect to spend a few minutes reviewing — not hours retyping. Blurry photos, glare, or handwriting reduce accuracy.
Does it translate the menu automatically?
Yes. The extraction and translation happen in one pass. Your final menu appears in the language you selected during signup. You can refine any translation manually — for example if "fish soup" should be "bouillabaisse" in French.
Which languages does the AI support?
The translation output supports English, French, Spanish, Bulgarian, Hebrew, Russian, and Greek. Gemini can also read source menus in many other languages — you don't need an English menu to start.
What if the AI gets a price wrong?
You review the extracted menu before publishing. Any price misreads are a one-click fix in the wizard. The AI is a starting point — it saves you hours of retyping, not hours of proofreading.
Why Gemini and not OpenAI?
Gemini 2.5 Flash is fast, cheap, and very good at structured extraction from menu images. The choice is a technical one — we benchmarked alternatives and picked the one that ships cleanest JSON for restaurant menus today. We may revisit as models evolve.