Chapter 19: Photo and Video Analysis

It is a feature of the Thread Diary (Chapter 18). You pick photos, or one video, from a thread, and the AI writes a short account of what is actually in them: a summary, a timeline, the people, objects and places it can see, and anything it is unsure about. The result is saved in the thread so you can read it again or ask about it.

Yes. Some screens in the app still call an analysis a "cook", for example the Past cooks list and the Delete this cook? message. Everything in this chapter is about that same feature.

No. The result is text, plus a row of small pictures from what you picked. It does not create a new video.

Open a Diary thread and tap the Mitra icon (the star). Choose Select items to analyze. The thread's items become tickable. Tick what you want and tap the sparkle icon in the top bar ("Cook selected"). A window titled Analyze this thread opens.

Yes. It sits under the Mitra icon inside a Diary thread, together with Past analyses and Ask a question (Chapter 15). It does not read your chats. That is Relation Analysis (Chapter 17).

No. The analysis runs on KuhuChat's servers, so you need a connection.

No. Saving things to a thread is free and instant (18.3). The AI runs only when you start an analysis or ask a question.

Photos, or exactly one video. You cannot mix photos and a video in the same analysis, and you cannot analyse two videos together.

Up to 15. On the thread screen you can tick up to 5 items. If you tick more you see "Select up to 5 items to cook." Once the Analyze this thread window is open, you can tap more photos there, up to 15.

The video becomes your only selection and everything else is un-ticked. While a video is selected, photos cannot be added. Un-tick the video first to go back to photos.

No. Only photos and videos are analysed. Text notes appear in the picker as grey tiles, but if you tick one the analysis stops with "Only images or a single video can be selected for analysis."

The video is longer than 2 minutes, or the app does not know its length, so a full analysis may not be possible (19.5).

The thread has nothing in it yet. Save some photos or videos into the thread first (18.3).

No. An analysis covers one thread at a time.

No. Select again next time.

An optional note to the AI, for example "Count how many people are in the photos" or "What is written on the board?". It is at the bottom of the window.

Up to 300 characters. Anything longer is cut off.

It follows your note only as far as the pictures and the video actually show. It does not invent things to satisfy a request, and it may say in "Worth noting" that the evidence does not support what you asked for.

No. Leave it empty for a general analysis.

1. Open the Diary thread and tap the Mitra icon.
2. Tap Select items to analyze.
3. Tick the photos, up to 5, and tap the sparkle icon.
4. In the Analyze this thread window, tap more photos if you want, up to 15.
5. Add a message if you like, then tap Analyze (the button shows the number of items).

The window shows "Looking through your photos and videos…" and then "Analyzing…".

Each photo is made much smaller on your phone and sent, through KuhuChat's server, to Google's Gemini AI. The AI then looks at the pictures directly and describes what it sees.

It describes what it can see, including how many people appear and what is clearly written in the picture. It does not guess names, relationships or the occasion.

Usually a few seconds to a minute. Keep the window open until it finishes.

Wait until the photo shows properly in the thread. A photo without a loaded picture cannot be analysed.

Open the Diary thread, tap the Mitra icon, choose Select items to analyze, tick one video and tap the sparkle icon. In the window tap Analyze video (full + audio). The window shows "Analyzing full video and audio…".

Up to 2 minutes. If the video is longer you see a message such as "This video is 2:30 — full analysis supports up to 2:00." Add a shorter clip to the thread and analyse that instead.

Yes. In a full video analysis the whole video file is sent, with its audio. What it hears is shown under "From voice notes" in the result details.

A very large video file (about 19 MB or more) cannot be sent whole. The app then switches by itself to a quicker method: it takes up to 15 pictures from the video and analyses those. This quicker method does not listen to the sound. It uses only one analysis from your daily limit.

It goes through the scene area by area, then checks again for people who are easy to miss, such as a child on someone's lap, someone asleep or leaning, someone partly hidden, or someone who appears only briefly. Counting can still be wrong, so the result carries a confidence label (19.6).

Only one video per analysis. Run a separate analysis for each video.

No. A video uses the whole analysis on its own.

A short summary, a row of small pictures (up to 8) from what you picked, and a Show details button. For a video you also see a line such as "Analyzed the full 1:20 video, including audio." and a confidence label.

Sections appear only when they have something to say:

• Timeline: what happened, in order.
• People: who or how many can be seen, without made-up names.
• Objects and Locations: things and places that are actually seen or stated.
• From voice notes: for a video, what was heard in its sound.
• From documents: text that was clearly readable.
• Worth noting: things the AI is not sure about.

It appears on video analyses. High confidence means the details are clear. Medium confidence means some details may be uncertain. Low confidence means please double-check the details below.

Most of the pictures gave the AI very little to work with, for example dark, blurry or empty photos with no caption. The summary will say there is not enough detail rather than make something up. Add captions or pick clearer photos and try again.

Yes. It can miss people or misread a scene. Treat the result as a helpful summary of your own pictures, not as proof, and look at the originals when something matters.

It is designed to describe only what the evidence shows. A thread might be a repair, an errand or a document trail as easily as a celebration, so it does not assume a happy story.

Yes. Tap Share. Your phone's share sheet opens with the summary text, and for a video the line about what was analysed. Pictures are not included.

It takes you back to the picker with nothing selected, so you can choose again and add a different message. It does not run again by itself, and a new run uses another analysis from your daily limit.

Yes. Open the analysis from Past analyses and tap Ask about this analysis (see 18.4).

Free accounts get 10 analyses in any rolling 24 hours, counted across all your threads together. Premium accounts have a higher limit.

A message: "You've used your analyses for today. Try again tomorrow, or upgrade for more."

Each analysis frees up 24 hours after you ran it. It is not a reset at midnight.

It may, because the limit is counted when the analysis starts. A video that is too long is refused before anything is counted.

No. It uses one.

No. Asking questions has its own limit of 50 questions per thread in 24 hours (18.4).

The app shows no charge for analyses.

In the thread, tap the Mitra icon and choose Past analyses. The screen is titled Past cooks. It shows the thread's 20 most recent analyses, newest first.

A short summary, up to 3 small pictures, and how many of the items had enough detail (for example "3/5 items").

Type in the Search past cooks… box at the top.

Tap its row. The saved result opens with the same Show details button.

Open it and tap Ask about this analysis. Your next question is answered from that analysis only.

Tap the bin icon on its row and confirm "Delete this cook?". It says "This removes it and its analysis permanently. This cannot be undone." Your original photos and videos stay in the thread.

You have not run an analysis on this thread yet.

They are saved inside the thread, so they follow the thread's privacy (see 18.1).

Only the photos or the video you select, and your optional message. Photos are shrunk first. For one video the whole file is sent, with its sound, unless it is too big, in which case up to 15 pictures from it are sent.

Google's Gemini, called by KuhuChat's server. The AI is not given your chats or other threads.

The written result, and the small pictures that were analysed, so that you can ask follow-up questions about that analysis later. Deleting the analysis removes both.

Avoid it. Anything you select leaves your phone for the AI service, so do not analyse pictures of passwords, bank details or ID papers.

Check your internet and tap Try again. Large videos can take a minute or more.

Nothing is selected yet. Tap at least one photo, or one video.

An item was deleted after you selected it. Close the window and select again.

The picture could not load. Check your internet and reopen the window.

That is the limit on the thread screen. Open the Analyze this thread window and tap more photos there, up to 15.

Run the analysis again with a message such as "Count everyone, including children on laps", or pick a clearer clip. Check the original video to be sure.