Better ai food analysis
Z
Zan C
The current description ai or photo ai modes are just quite poor at estimating the calories, content or even plain wrong - when when I specifically mentioned a food - it just doesn’t get added to the list. I think this is more of a model quality issue rather than system design. I’ve found models like gemini 3 pro are just amazing at this - even without any additional food descriptions needed.
Leah
marked this post as
open
Leah
Merged in a post:
Allow AI to generate on-the-fly analysis of food, rather than matching to a database
D
Daniel White
Matching items in the database is incredibly finicky. Every time I try to log a meal, I end up having to meticulously audit the items they match my meal to, correct the quantities, or find replacements.
There are two specific problems with this:
- It's incredibly time-consuming: The sheer amount of friction involved makes me legitimately reluctant to log food in the app at all.
- To be frank, I don't trust the accuracy. For prepackaged foods, scanning a barcode is easy, but for home cooked meals or when dining out, being forced to rely entirely on whatever generic options exist in the database leaves too much room for error. I can make my own custom food items, but this is also too time consuming to be done regularly.
Both of these problems become especially obvious when I go out to eat with my family. We usually eat family-style, meaning my plate is a sampling of small portions from several different dishes. Trying to accurately piece together a plate like that in Bevel's database is a bit of a nightmare. It takes way too much time to search for every single component, and even when I do, I don't trust the resulting data because the entries are often missing counts for things like fiber, caffeine, or added sugar.
In the future, I would love to see Bevel eliminate the reliance on the database altogether.
The ideal workflow would be natively integrating AI vision directly into the app. We should be able to snap a picture of a complex, family-style plate, add a brief voice or text description of what's on it, and let the on-app AI process the entire nutritional breakdown. No searching, no finicking with serving sizes from generic database brands, just seamless direct analysis and processing.
Leah
Hi Daniel White Bevel currently lets you log food by taking a photo and briefly describing it. Have you tried this feature, or is this what you were referring to?
If not, could you share a bit more detail about the feedback you have in mind? I’d be happy to help.
D
Daniel White
Leah sure, thanks for the reply.
Maybe an example would illustrate this best. Before I switched to my current workflow, I often had a problem where I snapped a photo and described my plate, and Bevel would get close, but be far off the mark in big ways. For example, I would attempt to log a breakfast with a photo and description like "15 raspberries, cup of coffee, half of a bagel (Dave's killer bread, epic everything) with butter, and core power protein shake". Bevel would return database entries for a bagel, a cup of raspberries, a coffee, and a random protein shake. However, it would have other unexpected things included too-- a high calorie raspberry dessert, or an extra half bagel hidden somewhere in the ingredients. Moreover, the generic bagel or protein shake would only have calories and macros--no fiber, sugar, etc.
Overall, the input methods are fantastic. I appreciate that you can take a photo, use a note, or combine these things for quick logging. It's what happens after the log is given to Bevel that is causing me an issue.
Leah
Daniel White: Understood! Will add this to the feedback board.
D
Doyun Koo
I’m not Zan C but I have the same issue. I have been personally using CHAT GPT to log my food and I noticed that the scanning ability at BEVEL is subpar than CHAT GPT’s. There are times when Bevel didn’t have information when I scanned the barcode. There were also times when I took photos of front and back so I can capture the nutritional information, but it didn’t register. It would be great if Bevel AI can integrate information that are scanned in the photo. I also noticed that the Bevel AI is not good at scanning ethnic foods that are not typically considered “American” food as well.
Amanda
Merged in a post:
Ai food recognition iteration/refinement
T
Tian
Sometime Bevel AI recognizes the food wrong. In such case I have to start over again. I would like to be able to iterate in the same conversation to correct the recognition result, instead of re-typing the whole food description . It saves time.
Amanda
marked this post as
waiting for user input
Hi Zan C, can you please share an example of this? Input pictures/description and the output?