How to use photo calorie tracking effectively (the 5 rules)

Photo-based calorie tracking is the fastest method we have. It's also the easiest to use poorly. Here are the five rules that separate effective photo tracking from "I took a picture and got nonsense."

Rule 1: shoot from above

The best angle for any photo log is straight down (90° to the table). Side angles confuse the depth estimation; oblique angles make portion sizes look wrong.

If you can't get directly above, get as close to top-down as possible.

The exception: tall foods like burgers or layered desserts where a side view shows the cross-section better. For these, a 45° angle works.

Rule 2: include a reference object

The AI is trying to estimate portion size from the image. The harder it is to gauge size, the worse the estimate.

Include something with known size in the frame:

• A standard fork, knife, or spoon
• A standard plate edge (most dinner plates are 10–11 inches)
• A wine glass or coffee mug
• Your hand (if you're comfortable)

A bowl floating in space against an unmarked tablecloth is the worst case for portion estimation. A bowl next to a fork on a regular dinner plate is the best case.

Rule 3: lighting matters more than you'd think

Bad lighting → bad food identification. Phones are good but not magic.

Best:

• Natural daylight
• Bright, neutral indoor light
• Restaurant table light (usually decent)

Worst:

• Dim mood lighting
• Yellow incandescent
• Strong colored light (blue tinted, red tinted)

If you're in a dim restaurant, turn on your phone's flashlight briefly to see the food clearly, then turn it off and shoot. (Flashlight while shooting can wash out colors.)

Rule 4: separate, don't stack

If your meal has multiple foods, lay them out so the AI can see each one. A plate with chicken, rice, and vegetables in three quadrants is much better than a bowl where everything is mixed.

For one-pot dishes (stew, curry, stir-fry) where separation isn't possible: photo log it as a single dish (the AI is trained on these as composite meals).

Rule 5: edit the result, don't just trust it

Photo tracking gets you to ~80% accuracy on the first pass. The remaining 20% comes from the user.

After the AI returns its estimate:

• Open the entry
• Check the items it identified — fix any wrong identifications
• Check the portion estimates — adjust up or down based on what you ate
• Save

Total time including editing: 15–30 seconds per meal.

The 5 seconds you save by skipping edits cost you accuracy you'll regret over a 90-day cut.

The common photo-tracking mistakes

Mistake 1: photographing the entire spread.

The AI tries to identify and portion every food in the frame. If you photograph the whole table at a dinner party, it'll log all of it as your meal. Photo only what's on your plate.

Mistake 2: photographing after you've eaten half.

The AI uses what's in frame as the portion. Photograph before eating, or photograph the original plate before serving yourself.

Mistake 3: photographing the menu instead of the food.

Some users photograph the restaurant menu thinking the AI will log the dish. It won't reliably — the menu doesn't tell the AI what you actually ordered or how it was prepared. Photograph the food.

Mistake 4: trusting the AI on completely unfamiliar dishes.

If the AI labels your bibimbap as "rice with vegetables and beef," that's roughly right but missing the gochujang and sesame oil. Edit the dish description for accuracy.

Mistake 5: never building a library of common meals.

If you eat the same breakfast every day, save it as a custom food after the first photo log. The next 100 logs take a single tap.

The photo + edit + save loop

The optimized workflow:

1. Photo (3 seconds)
2. Glance at AI estimate (5 seconds)
3. Adjust portions if needed (10 seconds)
4. Save (1 second)
5. If it's a repeat meal, save as custom (5 seconds)

Total: 15–25 seconds per meal, scaling down to 5 seconds for custom-saved repeats.

What about meals you can't photograph?

You can't photograph everything. Coffee in the dark of a 6 AM commute. A protein bar in the car. Snacks at a meeting.

For these, the workflow is:

• Voice-log: "tall coffee with oat milk"
• Quick-add: tap a saved favorite
• Type-search: "Quest bar"

Photo isn't the only mode. It's the best mode for full meals; quick-add is the best for snacks and drinks.

The CalorieScan AI photo modes

The app has three photo modes:

• Single dish: for plates, bowls, single items
• Multi-item: for combos like sandwich + chips + drink
• Recipe match: photograph a meal you've cooked before and the AI matches it to a saved recipe

Use the right mode for the meal.

The honest summary

Photo tracking is the future of consumer calorie tracking, but it's not yet hands-off. The user is part of the system.

Top-down. Reference object. Good light. Separate the foods. Edit the result.

Do those five things and your photo logs will be 90%+ accurate. Skip them and you'll be in the 60% range and frustrated.

The AI does the hard work. The 10 seconds of editing is the tax.