There are a number of apps available that use AI to estimate carbs based on a photo of what you’re planning to eat. “Estimate” is of course the operative word but I’m wondering if anyone has tried one and how well it worked? Recipes - especially for home cooked food - can vary so greatly I wonder if it’s even worth bothering🤔?
1 Like
Hi Dorie, no I haven’t but with the improvements of photo identification and the AI routines these little programs use get better and better. I used a similar app to identify a species of hedge in my front yard by taking a picture. Many of the carb estimating apps are pay/subscriptions but I think there are some free ones maybe try a few? Anyway good luck 
1 Like
I tried an app last year. It wasn’t useful but I’ll try again next year. I ran into some issues with the “take a picture” method though that made me think.
- By the time the food is in front of me to snap a pic it’s too late to bolus
- Getting out my phone, taking a pic and waiting for a response took way longer than I take to make a swag (sophisticated wild ass guess) about the number of carbs on my plate
- Consider Schrodinger’s burrito: there’s no way to know what’s in the burrito until you eat it. There are a lot of foods like that, salad is another one there’s no way to estimate how much is in there from a picture
Have you tried the meal picture logging in the Dexcom G7 app? It does AI ingredient detection, Dexcom says carb estimating is in development. Logging a pic is really great way to go because you can go back in Clarity or Glooko and see that you estimated x but next time x plus a bit more or less would be better.
So since there’s no way to tell what’s in the burrito it has everything and nothing simultaneously until you open (eat) it? 
thanks for the giggle Chris.
I experimented with Dexcom’s photo feature just to see if you had to use a “live” photo or if you could use a screenshot (you can). But with every one I tried I had to enter the carbs manually - which would be helpful to have in the future but as you say, may be more trouble than is worth for something you’re eating first time around.
If one DID provide a number at great I would find it useful as a baseline to bolus from, and once I figured out how hidden ingredients or manner of preparation factored in, I could adjust from there. I guess it’s a step up from an educated (?) guess.
1 Like
Full disclosure, I work on Carbie, a photo-carb app, so take this as an interested party rather than a neutral review. The objections raised here (opaque foods, speed, timing) are exactly the ones we hear most and haven’t fully solved either - no photo tool can see inside a burrito. Happy to hear what would actually make this kind of tool useful to you, since that’s the gap we’re trying to close.
1 Like
Since you’re here for business instead of pleasure can I ask how you got into working on a carb and calorie counting app targeted at people affected by diabetes?
Thanks for post, it reminds me I need to check out the logging in the G7 app now that I’m using it.
Sure, Chris. My son was diagnosed with Type 1 diabetes about four years ago, so carb counting quickly became part of our everyday life. At the same time, I’ve spent about 20 years working in software development, and it became pretty clear to me that there had to be a better way to make carb estimation easier, especially when eating out or dealing with meals where you don’t have nutrition labels.
That’s really where the idea for Carbie came from. It started as solving a problem we were dealing with ourselves, and eventually turned into something I thought could be useful for other people with diabetes and their families too.
Hi @amotylev welcome to Breakthrough T1D. “Necessity is the mother of invention”. There are already general population nutrition counters out there but what T1’s could use is insight. The fats in foods for example could have an impact on how carbs affect blood sugar. Pizza and ice cream are very different than an exact carbohydrate equivalent of white rice or breakfast cereal. I might suggest a “user in the loop” method to tune your counter or suggest when there is enough fat for a delayed and extended blood sugar event. Good luck!
1 Like
Thanks Joe, that’s a really good point. We’re thinking about Carbie in a similar “user in the loop” way rather than treating the first estimate as final.
Right now, users can adjust the detected food, change the portion/weight, and add or remove items if the AI didn’t interpret the meal correctly. That lets the person who actually knows the meal refine the estimate instead of relying blindly on the image analysis.
I also agree that two meals with the same carbohydrate count can behave very differently because of fat, protein, and meal composition. Giving people more context around that is an interesting direction for us, although anything that starts suggesting delayed glucose effects or insulin timing would need to be handled and validated very carefully.
I’d like to integrate this information into the Health Connect ecosystem I use for fitness, so asked Gemini for a synopsis. I’d opt for high accuracy, like SnapCalorie, but it doesn’t integrate with the systems I use.
Gemini:
| App | Key Features | Best For | Model/Platform |
|---|---|---|---|
| SnapCalorie | Uses depth sensors and volumetric AI to estimate portion volume; pulls verified USDA nutritional data. Supports photo + voice logging. | High portion accuracy & depth analysis | iOS / Android |
| Cal AI | Focuses on multi-angle photos for complex dishes and packaged meals. Provides detailed macro breakdown and target tracking. | General macro & calorie tracking | iOS / Android |
| SNAQ | Validated in clinical settings for carbohydrate and macro estimations. Integrates with Continuous Glucose Monitors (CGMs) like Dexcom/FreeStyle Libre. | Diabetes management & blood sugar insights | iOS / Android |
| Foodvisor | Scans the plate, identifies discrete components, and displays a visual overlay of ingredients and portion sizes. | Visual plate breakdowns & European/global recipes | iOS / Android |
| Calorie Mama (Azumio) | Built on deep learning models trained on diverse global cuisines. Offers rapid meal categorization. | Broad food classification & quick logging | iOS / Android |
| Nutrify | Extends tracking beyond macros to include micronutrients (vitamins, minerals) alongside visual food density analysis. | Whole-food & micronutrient detail | iOS / Android |
When asking about whihc one is best for ∂iabetics:
Gemini
For managing diabetes or pre-diabetes, SNAQ stands out as the overall best option among photo-recognition nutrition apps, with MyNetDiary (Diabetes Tracker) serving as a strong alternative.
Unlike general weight-loss or fitness apps, SNAQ was specifically designed as a clinical carb-counter.
Why SNAQ is the Top Choice for Diabetics
| Feature | How SNAQ Handles It | Advantage for Diabetes |
|---|---|---|
| Carb & Fiber Precision | Breaks down total carbohydrates, net carbs, dietary fiber, protein, and fat in under 10 seconds per photo. | Reduces carb estimation errors by ~38% compared to manual guessing. |
| CGM / Glucose Syncing | Connects natively with Dexcom (G6/G7/Stelo), FreeStyle Libre (2/3), and Eversense. | Overlays your logged photo directly onto your real-time glucose curve. |
| Health Connect Support | Reads and writes health metrics (activity, workouts, glucose logs) via Google Health Connect. | Keeps step data and exercise energy in sync with your glucose trends. |
| Time-in-Range Insights | Uses AI to analyze post-meal glucose spikes and spot recurring trends. | Helps pinpoint which specific ingredients or meal combinations trigger blood sugar spikes. |
Crucial Safety Note: While photo AI is helpful for pattern tracking and meal logging, never use AI photo estimates alone to calculate insulin bolus dosages. AI cannot reliably measure hidden sugars or exact fat/protein absorption rates. Always manually verify carb counts before taking insulin.
Alternative Options to Consider
-
MyNetDiary (Diabetes & Blood Sugar Tracker): Offers a specialized diabetes edition that includes barcode, photo-logging, and manual carb/insulin entries. Integrates well with Health Connect and blood glucose meters, though it focuses more on traditional logging than real-time CGM curve overlays.
-
Cronometer: Excellent for tracking precise micronutrients, fiber, and net carbs. Full Health Connect integration, but requires a manual or barcode-heavy logging workflow rather than camera-first AI recognition.
Key Recommendation
If you wear a Continuous Glucose Monitor (CGM) or want to see how specific meal photos directly correlate with your blood sugar responses, SNAQ offers the most practical setup.
Ah, But there is always a way. I’ve been in the pharmaceutical industry for 30 years. Instead of suggesting how to use insulin (regulated) you can use an unregulated synonym such as Glycemic Index (GI). Btw, GI is completely unregulate the FDA could care less. Good luck 
1 Like
That’s a powerful motivation. Here’s my ideas on how to improve the odds on carb guessing with an app.
- Start the lookup with a text prompt for what the food is. Image recognition should be an optional secondary way to refine the results. Allows for bolusing before the meal shows up.
- Use location data to check if a menu is available to increase relaibility and speed
- Eliminate the capture image button. Just start detection based on whatever the camera sees. The live feedback guides the user to capture the image that works best and reduces the feeling its slow by keeping the user interacting with the app.
- Use your document imaging wizardry to do the portion size calcs on device.
- There are nutrition databases for sale, e.g. Calorie King, combined with portion size detection this would be faster and more precise than using LLM for the whole lookup.
- Build Loop/Trio/AAPS integrations to enable a share button that sends the carb qty and using @Joe glycemic index idea an absorption duration time
- Using the Apple/Google Health as a database look up previous times I’ve has the same meal and the associated insulin/glucose/activity data
- Plan to license the Carbie technology stack to other pump and CGM companies
1 Like
Chris, this is incredibly thoughtful feedback. Thank you for taking the time to lay it out so clearly.
A lot of what you described lines up with areas we’ve been thinking about, especially making the experience faster, reducing friction around image capture, and using better context around the meal rather than treating every scan as an isolated image.
I really like the idea of combining image recognition with other signals like text input, location/menu data, previous meals, and potentially Health data. That feels much closer to how carb estimation actually works in real life — people usually know something about what they’re eating, even if they don’t know the exact carb count.
The Loop/Trio/AAPS integration idea is especially interesting. My son uses AAPS, and now we can sleep at night. If Carbie can eventually estimate carbs and then make it easy to pass that information into the tools people already use, that could make the workflow much more practical. The same goes for absorption-time estimates rather than treating every carb as identical.
Your point about nutrition databases is also well taken. We don’t necessarily want an LLM inventing the entire answer from scratch when structured nutrition data is available. A hybrid approach using recognized food, portion estimation, nutrition databases, and AI reasoning is probably the stronger long-term architecture.
And yes, licensing the technology to pump and CGM companies is something I think could become very interesting if we can prove the product and the underlying estimation workflow at scale.
I’m going to save this list because there are several ideas here that are genuinely worth exploring.
1 Like