Video for AI chats

Why this tool exists

Many AI chats do not accept a video as an attachment, and even where one is accepted, tight duration and size limits apply; images and text, on the other hand, work in every chat. This tool turns a video on your device into frame sheets (grids in a single image, sized for chats, with the number and the time printed under every frame) and into a context text that tells the AI how to read them. The tool lays the video out, it does not interpret it: the actual reading is done by the model you send the sheets to.

How to use it in three steps

Load the video and let it run: under 20 minutes the tool starts on its own with the recommended settings. Then 1) download the sheets, 2) copy the context text, 3) in the chat attach the images and paste the text as your first message, finishing the question at the bottom. On a phone you can also long-press a sheet preview to save it. For long videos, the From and To fields in Options concentrate the frames on the stretch you care about.

The two ways of picking frames

“At regular intervals” takes one frame at the middle of each slice, all equally spaced: the right choice for continuous footage (a walk, a fixed shot, a screen recording). “At scene changes” runs a scan first and tries to place frames where the picture changes abruptly: it works best on edited videos (sports matches with replays, lectures with slides, vlogs). The detection is a heuristic on the difference between nearby frames, not professional shot detection: on static videos it finds nothing, says so, and falls back to a nearly uniform coverage by itself. Keep one typical case in mind: a vibration or flicker is invisible in any still frame; there, pick Detailed and describe the symptom in words in your question.

What the context text is for

It tells the AI what the images are and how to read them (left to right, row by row), repeats in writing the number and time of every frame (if the chat recompresses the images and the printed labels degrade, the map survives) and states the limits of the package, so the model does not invent what cannot be seen between two frames. The “My question:” line is left open on purpose: you finish it in the chat. You can build the text in Italian, English or Spanish, in the language you are chatting in, without redoing the sheets.

File size and attachments

Attachment limits change from chat to chat and over time, so you will not find any provider numbers here: you find the actual file size of each sheet on its label and three size targets (1, 2 or 4 MB) in Options. When a sheet exceeds the target the tool steps the quality down and then shrinks the sheet, and it says so when it cannot go lower. Sheets are 1568 pixels wide because chats resample big images: extra pixels would just be extra bytes. Always review the sheets before sending them: a sheet carries no EXIF data or GPS position, but everything visible in the frames stays visible.

What it does not do (and the honest alternatives)

It does not transcribe audio: that would mean downloading a speech model tens of MB in size on every visit, out of scale for this page. If the speech matters, paste the subtitles into the Options field (they end up in the context text), or use your phone keyboard dictation while replaying the video; some assistants also accept audio files as attachments, and in that case send the audio along with the sheets. It does not send anything to the AI: it produces files and text, attaching them is up to you. It does not guarantee exact frame number N (a web page asks for times, not indexes) and it does not show the movement between two frames: a gesture that falls in between does not exist in the package, and the honest remedy is narrowing From and To on the stretch that matters. Need a single frame at full resolution? Use Extract a frame from a video. The video will not open (HEVC/H.265, many MKVs)? Convert it first with Convert video.