Home/Tools/Estimate Calories

Estimate Calories from a Picture: How Good Is AI?

Paul Hummel Last reviewed October 6, 2026 3 min read
The photo is sent encrypted to OpenAI for analysis and is not stored by us. More in the privacy policy.
Analyzing…

0kcal
0 gCarbs0 %
0 gProtein0 %
0 gFat0 %
Servings
1
Ingredients

    Track every meal with daily targets and history: in the GymLog AI app.
    You have used your free scans.

    Keep scanning in the GymLog AI app – with daily targets and history.

    Get the app for free →
    Quick answer

    According to the systematic review by Shonkoff et al. (2023), AI methods for estimating calories from photos align with the accuracy of human estimates. Average relative errors ranged from 0.10 to 38.3 percent depending on the study.

    AI vs. human estimates

    Shonkoff et al. (2023) describe human error in estimating food intake as a major source of bias in nutrition research. Reviewing 52 studies, they concluded that AI methods align with – and have the potential to exceed – the accuracy of human estimates.

    How to get a better estimate

    Errors were lower for single, simple foods than for complex dishes. It helps to arrange components visibly, shoot from above in good light and add hidden ingredients like oil, butter or dressing after the scan.

    A reference object in the picture, such as a fork next to the plate, can help the AI judge portion size.

    Track every meal in seconds

    GymLog AI logs meals via photo, voice or chat – with daily targets for calories and macros.

    Download on the App StoreGet it on Google Play

    Sources

    1. Shonkoff, E., Cara, K. C., Pei, X. A., et al. (2023). AI-based digital image dietary assessment methods compared to humans and ground truth: a systematic review. Annals of Medicine, 55(2), 2273497. PubMed
    Results are AI estimates and may deviate considerably. They are not a substitute for nutritional advice. If you have a medical condition, allergies or an eating disorder, please consult a professional.