AI Meal Planning from a Fridge Photo — The Micro-Habit That Actually Saves Sundays

Why meal planning fails every week

The problem with meal planning isn't motivation. It's that the planning itself takes effort — browsing recipes, checking what you have, matching ingredients, thinking about variety across seven days. By the time you've decided what to cook Monday, you're tired of the process and just repeat last week's rotation.

What AI is actually doing here is removing the decision-making step. You're not asking it to cook for you or even tell you what you should eat. You're asking it to look at what you already have and do the matching work that used to take 30 minutes of mental overhead in about 30 seconds of processing.

Once the AI does the matching, your job is just to look at the output and say yes or no. That's a much smaller cognitive task. And that's why it sticks — the habit is "take a photo", not "plan meals for a week".

The actual method — photo, prompt, plan

The setup is genuinely simple. Here's the sequence I've used every Sunday evening for the past three months:

The base prompt — copy this exactly

This is the prompt I've refined over a dozen Sunday sessions. The specifics matter — vague prompts give vague meal lists.

Here's a photo of my fridge (and pantry if I included a second image).
Based only on what you can see, suggest 5 dinner ideas and 3 lunch ideas
for the week.

Rules:
- Use only ingredients visible in the photo. Don't suggest dishes that
  need things I clearly don't have.
- Keep each meal under 30 minutes of cooking time.
- Flag anything that needs to be used soon (looks close to expiry).
- If you can't identify something clearly, ask me what it is — don't guess.
- I'm cooking for 2 people.

Format each suggestion as:
[Meal name] — [2-sentence description of what it involves and roughly how long]

Once you've used this a couple of times, you'll start adapting it. I added a line for "one of these should be a no-cook option" after a particularly hot week in April. You can add dietary restrictions, cooking skill level, or specific cuisine preferences. The base prompt above is the starting point, not the final form.

For a more structured weekly plan

If you want a full Monday-to-Sunday breakdown rather than a list of suggestions, this variant works well:

Based on the fridge photo, create a simple weekly meal plan (dinner only).
Monday to Friday: quick meals under 30 minutes.
Saturday: something slightly more involved if the ingredients allow.
Sunday: use up whatever is left before it spoils.

For each day list:
- Meal name
- Key ingredients used (from the photo)
- Approximate time

Flag anything you're unsure about in the photo.

ChatGPT vs Claude vs Whisk — which one actually works

I've tested this consistently across three tools. Here's my honest read after months of Sunday use:

My actual recommendation: Claude for Indian households, ChatGPT if you cook mostly Western food. The difference comes down to how each model handles uncertainty. Claude tends to ask "I can see what looks like a container of something orange — is this dal or a curry base?" ChatGPT will call it tomato soup and build a recipe around that assumption.

Both are good. For my kitchen, Claude's caution has led to better outputs more consistently.

A note for Indian kitchens specifically

Most AI food tools are trained heavily on Western food datasets. A fully stocked Indian fridge — pressure cooker dal in a steel container, green chillies in a bag, half a bunch of coriander, a block of paneer, assorted dabbas with unlabelled contents — genuinely confuses them.

The workaround that actually helps: add a sentence describing what's in your pantry in text, even while attaching the photo. Something like: "My pantry always has: basmati rice, toor dal, chana dal, atta, and a full spice box." This fills in the gaps that even good image recognition will miss.

Add this line to the prompt and the suggestions become dramatically more accurate:

My pantry staples (always available, not in the photo):
basmati rice, toor dal, chana dal, atta, mustard seeds, cumin,
turmeric, red chilli powder, coriander powder, garam masala, ghee.
Suggest Indian meals where possible — I cook primarily North Indian food.

When it doesn't work — be honest with yourself

Three situations where this habit consistently breaks down for me:

When the fridge is genuinely empty. If you're down to half an onion, two eggs, and a lot of condiments, AI will try its best but the suggestions will feel forced. The habit works when you have ingredients — it's not a substitute for actually buying food.

When you're cooking for specific dietary requirements or medical restrictions. AI meal suggestions are general. If someone at home has a specific health condition that affects what they can eat, don't rely on AI to automatically know and account for that. You need to explicitly state constraints every single time, and even then, verify.

When you're very tired and just want to be told exactly one thing to cook. The output is a list of suggestions. If you're in decision fatigue mode, a list still requires a choice. In those moments I ask a simpler follow-up: "Pick the easiest one from that list and give me step-by-step instructions." That helps.

What about Snap Calorie?

Snap Calorie is good at what it actually does — estimating calories from a photo of a plated meal. It's not designed for this use case (ingredient-level fridge planning). I tried it for a week out of curiosity and the results were predictably not useful — it kept trying to estimate the nutritional content of my raw vegetables rather than suggest what to do with them.

Worth knowing it exists if you're tracking macros, but for the planning habit we're building here, it's the wrong tool.

The habit itself — what makes it stick

The reason this one has lasted for me when other meal planning attempts haven't is the trigger. I open the fridge Sunday evening to see what's for dinner — a thing I do anyway. I've just added: take a photo before closing the door. That's the entire habit addition. Everything else flows from that one moment.

After four weeks, I added one more step: I paste the weekly plan into a WhatsApp note I share with my wife. Two minutes total. We both know what's for dinner each day without a conversation. The mental overhead is gone — not because I have more discipline, but because the decision was made Sunday and doesn't need to be made again Wednesday night.

Verdict