Turn screenshots of a group chat into what everyone decided.
Your screenshots are sent to a server so a vision model can read the conversation. They are not stored. These are other people's messages — if that is wrong for a particular chat, don't use this one.
Up to six, oldest first. Screenshot, scroll, screenshot again — overlap is fine and helps. Any app: iMessage, WhatsApp, Instagram, Discord.
Reading the conversation…
This sentence is the only thing sent to the model that draws.
Replaces the names it recognised, everywhere they appear. A nickname it never saw as a sender will stay — fix those below.
Your screenshots are sent to a server so a vision model can read the conversation in them. They are not stored. The screenshots are held in memory only while the conversation is being read — a second or two — and are never written to disk, logged, or used to train anything. The card is built in this browser tab. If you are signed in and ask for generated art, that request carries one sentence describing a place — shown on the page, and editable, before you press the button — and the list of names already read out of your screenshots, which is used only to refuse a description that names one of them. Only the sentence reaches the model that draws. Your messages, your screenshots and the card itself are not part of it.
Drop screenshots of a group chat and get back a card you can paste into the thread: what everyone decided, who said they would do what, the dates and times that came up, what actually happened, and the questions nobody ever answered. Up to six screenshots, because a conversation is longer than one screen.
It is not a summary. A summary of a group chat is another wall of text, and the whole reason you are here is that you already have one. Five sections, short lines, names attached to the things people took on — the shape of a thing somebody can act on rather than read.
The card comes out as a picture, sized for a story: a headline, the emoji the conversation itself was using, and the facts on top of a colour picked from how the chat was going. Eight looks to choose between, or press Surprise me. Changing the look costs nothing and happens instantly — the card is drawn on your device, so nothing is sent again when you pick a different one.
Most group chats do not plan anything, so the card does not assume yours did. An argument that reached a conclusion, a score, a bit of news somebody passed on — that goes under what happened, and it is often the only part with anything in it. A chat that settled nothing gets a card that says so, rather than an error blaming your screenshots.
The screenshots overlap, and that is expected: screenshot to the bottom of the screen, scroll, screenshot again. A message caught twice is counted once. Put them in the order they happened and the card follows the conversation.
Reading a conversation out of pictures needs a vision model far too large to run in a browser tab, so the screenshots are sent to a server, read, and the card comes back for you to correct before you save it. They are held in memory for the second or two that takes, never written to disk and never kept.
The plan changed three times overnight and the useful part is four lines somewhere in the middle. Screenshot the run and find out where it landed.
Somebody said they would book it, somebody else said they would sort the money, and by Friday nobody remembers which was which. The card puts a name next to each one.
Paste the card back into the chat the day before. It is the fastest way to find out that the one question everybody skipped is still not answered.
Somebody sends you screenshots of the last two days rather than adding you to the thread. This turns them into the decisions without you reading the whole argument.
Take up to six, scrolling between them. Overlap is fine and helps — it is how the tool knows the screens join up.
They are numbered as they land, and you can move any of them left or right. That order is the conversation's order, so it is worth a glance before you send.
Every line is editable. A misread name or a wrong time is faster to correct here than to argue about after you have pasted it.
Eight looks, or press Surprise me. The card is saved as a picture — drop it back into the chat, or post it to a story.
Yes — this is one of the few tools here that has to send something. Reading a conversation out of a picture needs a vision model that cannot run in a browser tab. The screenshots are read and discarded: held in memory for the second or two the reading takes, never written to disk, never stored, and never used to train anything. If that is wrong for a particular conversation, do not use this tool for it.
That is a judgement about your conversation and not one this page can make for you. What the tool does is keep the card to what was settled rather than what was said: it does not quote messages, describe anybody, or comment on how the conversation went. There is also a switch that replaces every name with an initial before you save the picture.
It replaces the names the tool recognised as people in the conversation, everywhere they appear on the card — including inside the open questions. A nickname it never saw as a sender will stay put, which is the honest limit of it. Every line is editable, so the fix for a missed one is to correct it before you save.
Most often the screenshots did not cover it — a gap between two of them, or a reply to something further up. The card says what it could not see, and that note is always shown even when it is empty, because a card that quietly leaves it out looks identical to one that read the whole thing.
Because they said maybe. An offer nobody answered goes in the open questions instead. Inventing a commitment for somebody who never made one is the one mistake this card cannot make, since it is going back into the conversation the people in it can read.
Any of them. It reads the picture, not the app, so iMessage, WhatsApp, Instagram, Discord, Messenger, Signal and Slack all work the same way. Dark mode is fine too.
Both are read out of the conversation. The emoji are the ones the chat itself was using — in the messages or as reactions — and fall back to ones that fit the subject when it used none. The headline is the one sentence on the card that is written rather than quoted, and it has to be true from the screenshots: it cannot announce something nobody agreed, sharpen a maybe into a yes, or make a joke about anybody. Both are editable before you save.
If you are signed in you can have a background drawn for the card. The only thing that reaches the model that draws it is one sentence describing a place — something like “a floodlit pitch, wet grass, late evening”. It is written for you from what the conversation was about, and it is shown on this page where you can read it, rewrite it, or clear it before you press anything. Your messages, the screenshots and the card do not go anywhere near it. A brief that names somebody in the conversation is refused and a generic scene is used instead — which is why that request also carries the names already read out of your screenshots, and nothing is done with them except that check.
Cost, and nothing else. Reading your screenshots costs a fraction of a penny; drawing a picture costs about a hundred times that, so it is capped per account and capped again across the whole site each day. Everything else here works signed out, and the eight themes cost nothing and are instant — they are not a lesser option, just a free one.
The card is saved as an image, which is what pastes cleanly back into a chat. The lines are all on this page as text while you are editing them, so you can copy any of them from here before you save.