Room photos can help you remember what you own, but they are hard to use as an organised list. You still need names, locations, and details that may not be visible in the image.
OGAD (Off Grid AI Desktop) can help you draft a home inventory from photos using a local vision model. Ask it to list visible items, check the result yourself, and add confirmed details from your notes. After model setup, image analysis can run on your computer without sending the photos to a cloud AI provider.
| Download OGAD | Desktop releases |
What would you like to do with Off Grid AI?
Have a feature or use case you would like us to support? Tell us what you want to do and which device you use.
Write to support@offgridmobileai.co, join our Slack community, or talk to us on Reddit.
Use the model as a first-pass assistant. A photograph cannot establish everything you own, and visual recognition can miss or misidentify objects. Your checked inventory is the record you maintain.
What can a photo tell you?
A clear photo can show visible objects and their location in the room. It may show a label or distinctive feature. It usually cannot establish a hidden serial number, purchase date, exact model, condition, or value.
Suppose you are organising a home office before a move. The photo shows a desk, a chair, two monitors, and several accessories. A useful draft separates those visible items from details you need to check in person.
| Inventory field | Where the information should come from |
|---|---|
| Item description | Photo, checked by you |
| Room or location | Your label for the photo |
| Brand or model | A readable label or your records |
| Serial number | A direct checked record |
| Purchase detail | Your receipt or other documentation |
| Uncertainty | Anything the photo cannot establish |
Do not ask the model to fill every blank. A visible gap is better than a convincing but incorrect identifier.
What do you need in OGAD?
Install OGAD and download a compatible local vision model with its required files. A text-only model cannot analyse photographs simply because an image is attached. Complete app, model, and runtime downloads before working offline.
The workflow uses free core vision chat on supported Mac and Windows computers. 0.0.54-beta.108 also provides Linux beta packages.
Choose a small set of clear photos saved locally. Begin with one room and a few items. A wide shot helps with location, while a close-up can help you read a label. Avoid expecting a tiny object at the back of a room to be recognised reliably.
Keep original photos and your existing notes. The inventory should refer to those sources so you can check it later.
How do you create the first room list?
Select a local model that supports image input, attach one photo, and ask only for what is visible. Keep the first request narrow enough to check quickly.
- In Models, download and select a compatible local vision model.
- Open a new chat and use the + menu’s Add image control.
- Choose a clear room photo.
- Give the room a neutral label in your prompt.
- Ask for a draft list with uncertainty kept explicit.
The chat implementation passes image files to the multimodal model. This is image analysis, not automatic access to the rest of your photo library.
Try:
Draft an inventory of visible items in this home-office photo. Use ordinary descriptions and approximate positions. Do not guess exact brands, models, serial numbers, prices, or items hidden from view. Mark uncertain identifications as uncertain. Treat this as a draft for me to check.
How do you avoid duplicate items across photos?
Review one room at a time and keep a photo label beside each draft entry. When two images show the same object, decide whether to merge the entries yourself. Similar objects may be separate items, and the model can confuse them.
For the office example, a monitor visible in a wide shot and a close-up should not become two monitors in the final list. Conversely, two similar monitors should not be merged merely because they look alike.
Give items your own simple identifiers after checking them, such as office-monitor-01. Use those identifiers in later notes and close-up filenames.
Ask the model to propose possible duplicates, not to delete entries automatically:
Compare these draft lists and flag entries that may refer to the same item. Explain the reason for each suggestion. Keep uncertain matches separate for my review.
How should you add labels and purchase information?
Read important identifiers directly and compare them with the image. A local vision model can help transcribe visible text, but small labels, reflections, and blur can produce errors.
Use a close-up request such as:
Read only the visible label text. Preserve characters as shown and mark unclear parts. Do not complete a model or serial number from a likely product name.
Check the result character by character before adding it to the inventory. Copy purchase details from your records and keep the source. Do not use an AI guess as an appraisal or evidence of ownership.
If you want an inventory for an insurer or another organisation, check its requirements separately. This article creates a general record and does not establish that the resulting document meets a particular evidence standard.
How do you turn checked entries into a useful document?
Save a plain table with the fields you actually need. You can ask the local model to format your reviewed entries, then copy the table into your normal spreadsheet or document editor.
Format these checked entries into an inventory table with item ID, room, description, confirmed identifiers, source photo, and notes. Preserve blank fields. Do not add values or change identifiers.
Review the output after copying it. Confirm that no rows were merged or dropped. Keep the source photos and notes with the inventory through your own file-management process.
The chat workflow does not automatically create a verified insurance schedule, synchronise an inventory database, or track when an item leaves your home.
What should you check before calling it complete?
Walk through the room and compare the list with the items. Photos may omit cupboards, drawers, or objects outside the frame. Add those deliberately if they belong in your inventory.
| Issue | Useful correction |
|---|---|
| An object is misidentified | Replace the label with your checked description |
| Two views create duplicates | Merge only after confirming they show the same item |
| A serial number is uncertain | Read the physical label or leave it blank |
| Small items are missing | Take a clearer photo or add them manually |
| The list includes guessed values | Remove them unless you have a separate verified source |
Inventory one room first
Download OGAD, load a local vision model, and analyse one clear photo. Check the visible items, add confirmed details, and save a table linked to your source images.
Keep the vision model local for the processing described here. Initial downloads and later sharing or backup use their own connections. The useful result is an organised record you have checked, built from material already on your computer.