Protealpes

AI tool

Analyze meal calories from text or a photo

Upload a plate photo, describe what you ate, or both — then get calories and macros mapped to food data.

Drop a photo or click to upload

JPG, PNG, or WebP · or switch to Live scan above

Optional. One plate, well lit, from above if you can. Upload a file or use Live scan.

Optional if you uploaded a photo. Add oil, drinks, brands, and amounts the camera cannot see.

Results stay on this page. Photos are not stored.

Totals are educational estimates. They are not a medical diet analysis or an allergen guarantee.

What this tool does

You upload a plate photo, type what was on the plate, or both. The model splits that into items, matches them to a USDA-derived table when it can, and adds calories, protein, carbs, fat, and fiber. Unmatched foods stay marked as estimates. It does not watch you eat and it does not replace a weighed recipe.

How it works

A photo is a noisy ingredient list; a sentence is a noisy one too. Vision names what it can see. Language fills oil, drinks, and brands the camera misses. Arithmetic is not the messy step: matched rows use ordinary serving math. Unseen fat is still the usual miss — write it if you know it.

How to use it

  1. Add the meal. Upload a plate photo, type the foods and amounts, or both. Mention oil, drinks, and brands in the notes.
  2. Analyze. The model splits the plate into items and matches USDA-derived rows when the names are ordinary.
  3. Edit the ledger. Fix portions, add hidden fats, and remove extras before you treat the total as a log.

How to interpret the result

Treat the total as a draft. If the rice looked shiny, add oil. If the chicken was a restaurant breast, it may include injected brine and a butter finish the description missed. Compare the protein grams to your daily target, not to a stranger’s bowl.

Example

“200 g cooked chicken breast, 180 g cooked white rice, 1 cup broccoli, 2 tsp olive oil” should land near 530–580 kcal with about 50 g protein, 55 g carbohydrate, and 12–14 g fat on typical USDA rows. Leave out the oil and the same plate prints ~80 kcal light — that is the usual homemade miss.

Common mistakes

  • Writing “healthy bowl” with no weights or fats.
  • Forgetting drinks, bread on the side, and the cook’s oil.
  • Trusting a 612 kcal total as if it were a bomb calorimeter.
  • Logging a family-size tray as one serving.

When this estimate may be off

  • The meal is a restaurant special with no public recipe.
  • Brand sauces and bakery items that the table cannot match.
  • You described a feeling (“light lunch”) instead of foods.
  • Hidden fats (mayo under the bun, butter on the steak) stay unsaid.

Frequently asked questions

Is this as accurate as a food scale?

No. A scale plus a saved recipe is tighter. This page is for plates you already ate or cooked by eye.

Where do the calories come from?

Matched items use USDA-derived rows. Unmatched items are model estimates and should read as such. The site does not invent a fake precise ledger for mystery food.

Should I name brands?

Yes when you know them. “Chobani 0% 170 g” beats “yogurt.” The matcher still may land on a generic cup.

Can I log a whole day here?

You can paste several meals, but error stacks. One plate at a time is cleaner. Use the recipe tool for a pot you will repeat.

Does it count cooking oil I did not mention?

Only if you write it. Nonstick spray is small. A restaurant sauté is not.

Can I analyze a meal from a photo?

Yes. Upload a plate photo on this page, add notes for oil or drinks the camera cannot see, then analyze. Matched foods still use the USDA-derived table. Unmatched rows stay estimated.

Should I use both a photo and a description?

When you can. The photo names what is visible. Notes catch cooking fat, drinks, second helpings, and brand names. That combination beats either input alone.

What photo works best?

One plate, overhead if you can, in ordinary indoor light. JPG, PNG, or WebP. A dark restaurant shot or a pile of mixed sauce will produce more estimated rows.

Related tools

Related guides

How Protealpes calculates