The photo AI is fast but not perfect. Sometimes it labels a dish wrong, invents a side that wasn't there, or estimates the portion at twice what you actually ate. You have four ways to fix this; here's when each one is the right move.
1. Edit the result (most common, fastest)
If the AI got mostly the right items but the portions or names are off, just edit the result before saving. Open the item, adjust the portion (or switch units), rename it, or remove items that weren't on the plate.
This is the right move when:
- The AI got the right dish, wrong portion (e.g. "1 cup of rice" when you ate half a cup).
- The AI invented an item (e.g. listed "side salad" when there wasn't one).
- The AI used the wrong name for the right thing (e.g. "chorizo" when it was andouille).
Full editor walkthrough: Edit or correct a logged meal.
2. Re-shoot the photo
If the AI got the items completely wrong — labelling chicken as fish, missing half the plate, calling a curry a soup — the photo itself was probably the problem. Cancel, recompose, and try again. Most often the fix is one of these:
- Top-down angle instead of a steep side angle.
- Brighter light — natural or a brighter overhead source.
- Everything in frame — if a side dish was cropped out, include it.
For the full set of capture tips: Log a meal from a photo.
3. Type the meal name instead
When the photo flow keeps missing on the same dish — usually an obscure home-cooked meal, a regional dish outside the AI's training set, or a heavily mixed composite — skip the camera and type the name directly.
See Log a meal by typing the name for naming patterns that work best.
4. Send feedback (to improve the model)
If the AI is consistently wrong on a dish that other apps would also miss — particularly for shared-meal cuisines and home cooking — shake your phone to open the feedback form. Describe what you logged, what the AI said, and what was actually on the plate.
This won't fix your specific meal in the moment (use one of the other three options for that), but it directly informs how we tune the AI over time. The shared-meal-cuisine gap is the problem BeforeIBite was built to address, and your real-world examples are the highest-quality signal we have.
Which option to pick
| The AI got… | Use… |
|---|---|
| Right items, wrong portions | Edit |
| Right items, wrong names | Edit |
| Some right, some wrong | Edit (remove the wrong ones, add the missing ones) |
| Completely off (different dish entirely) | Re-shoot |
| Photo wasn't the problem — it's the dish | Type the name |
| Consistently wrong on the same dish | Type it AND send feedback |
For the broader picture on what to expect from AI accuracy, see How accurate is BeforeIBite's AI?.