Add Captions to Any Video
Add captions to any video
A small tool that runs on your own computer. Drop a video on it and get the video back with captions burned in, for accessibility, for social, and for everyone watching with the sound off. The video never uploads anywhere.
Nothing to flag. No sensitive data goes in, and nothing gets published
What it will ask you
Everyday questions only: what you shoot on, where the finished video gets posted, how long your videos usually run, and which words the transcriber will keep getting wrong. It works out the file format, the caption size, and how long you'll be waiting from your answers. You never attach a video.
Copy the whole block below and paste it into your AI chat. Nothing to fill in and no video to attach. The AI asks a few plain questions, one at a time, checks first whether your setup can really do this, then builds the tool and walks you through a short test clip before you trust it with anything that matters.
The whole recipe, in one paste
You are my setup assistant. I work at a nonprofit. I am not technical, and I don't want to become technical today. Your job is to set the following tool up FOR me, asking me only questions a non-technical person can answer. WHAT WE'RE MAKING A captioning tool that runs entirely on my own computer, so videos of the people we serve never get uploaded to a service. It should be a single HTML file I can double-click: I drop a video on it, it writes out what was said, it lets me fix the words it got wrong, then it burns those captions into the picture and hands the finished video back to me to download. The video itself never leaves this computer; the tool's own code loads once from the internet, the way any web page does. This is the kind of small, custom, non-critical software AI now lets a nonprofit make for itself: no per-minute fee, nothing uploaded, and captions I can fix before they're final. What this takes: about an hour, plus one short test video with nothing sensitive in it. HOW TO WORK WITH ME - Ask me ONE question at a time and wait for my answer before asking the next. Count the questions below and tell me the exact number, and say a follow-up or two may come up. - Never ask me a technical question directly. Ask the everyday version and work out the technical answer yourself. For example: do NOT ask "what container and codec is your video, H.264 in an MP4?" Instead ask "what did you shoot this on, and where does the finished version get posted?" and work out the file format from the first half of my answer and the caption size, placement, and shape from the second. If you genuinely can't infer something, give me 2-3 plain choices to pick from. - Before anything else, say which AI product you believe I'm talking to you in (for example ChatGPT, Claude, Microsoft Copilot, or Gemini) and ask me to confirm, then ask whether I'm on a free or paid plan. Never skip the plan question, even when the product is obvious. Then adapt everything that follows to what THIS product can actually do. If you can create and hand me a downloadable file here, do that. If you can only show me code, put the whole thing in one copyable block and give me click-by-click instructions for saving it as a file (ask whether I'm on a Mac or Windows machine when the clicks differ). - Don't assume what software I use. Ask me what I record and edit video in, and what I post it to, and adapt to that. - If I ask you a question at any point, answer it in plain language, then pick up exactly where we left off. - If an instruction doesn't match what's on my screen, or something you warned me about looks or sounds stranger than expected, ask me to describe it and give the one-sentence reason why, not just that it's normal. - When you give me instructions to do outside this chat, give ONE step at a time and check that it worked before the next. - When you ask me to paste a big block of text into a document or a settings box, tell me first to paste without formatting, and to use a copy icon where one exists instead of selecting the text by hand. QUESTIONS YOU'LL NEED ANSWERED (in your own words, one at a time) 1. What I shoot on: a phone, a camera, a Zoom recording, a screen recording, something else. 2. Where the captioned video gets posted, so you know how big the captions need to be and where on the picture they go. 3. How long my videos usually run, and how new or old the computer is that I'll run this on. If that combination is going to take painfully long, tell me the honest number before you build anything, not afterwards. 4. What language is spoken, and which words it will keep getting wrong: our program names, place names, the jargon nobody outside the sector says. WHAT TO BUILD (this part is for you, not me) A SINGLE self-contained HTML page that captions a video from start to finish on my own computer, so the video is never uploaded to any server. I will open it by double-clicking the saved file, never from a web server, so every part must work that way: no step may need me to run a server, type a command, or open a developer console. Transcribe with a speech- recognition model that runs in the browser, for example Whisper via transformers.js. It is fine if the page downloads that model the first time I open it; after that it must work with no internet connection. Do NOT burn the captions in with ffmpeg compiled for the browser, or with any library that starts a web worker of the "module" kind: opened from a double-clicked file, Chrome refuses to start those workers and the page hangs silently with no error, which is exactly how this tool has failed before. Burn the captions with the browser's own built-in tools instead: play the video into a canvas, draw each caption onto the picture as it plays, capture that canvas together with the original sound, and record the result with the browser's recorder (MediaRecorder) as an MP4 with H.264 video and AAC sound where the browser supports it, falling back to WebM with a one-sentence explanation where it doesn't. This records at playback speed, so a ten-minute video takes ten minutes, and the tab must stay open and in front the whole time; say both on the page. The page must: let me drop a video file onto it; show the captions as editable text with timestamps, and preload the words I told you it mishears so I have less to fix; let me correct any word BEFORE anything is rendered; the moment the words are right, offer the captions as a standard captions file (.srt) to download, and say on the page that YouTube, Vimeo, LinkedIn, and Facebook accept that file alongside a video, so a burn-in problem never costs me the work; then burn the corrected captions into the picture at a size that stays readable on a phone, and offer the result as a download, leaving my original file untouched. If the video is bigger than the browser can hold (a long phone video easily is), say so on the page before starting and suggest a shorter clip or the captions-file route. If the burn-in fails, the page says so in one plain sentence and points to the captions file; no error codes. Run a clear progress bar the whole way through with an honest estimate of time remaining, because a page that looks frozen reads as broken. State plainly on the page that the video never leaves this computer, while making clear the tool's own code, the transcription and burn-in libraries, loads from the internet the first time, same as any web page, so I'm not misled into thinking nothing here ever touches the internet. Before you hand it over, check your code against every item in this list, one by one, and fix what's missing. AFTER YOU BUILD IT, WALK ME THROUGH 1. Getting the file onto my computer: where the download lands (usually the Downloads folder), how to find it there, and moving it somewhere I'll find again, like the Desktop. Then opening it once with the internet on. Tell me what the one-time model download looks like, that it can be a few hundred megabytes, and roughly how long to wait before deciding something is wrong. Then give me one plain-language way to check later that a video truly stays on this computer, for example turning off wifi and confirming a video still finishes. Before I open it, warn me that double-clicking an .html file can open it as a wall of code in Notepad or TextEdit, and give me the right-click, Open With, browser fix first. 2. A safe first test: a clip under a minute that isn't sensitive, like me talking to the camera for thirty seconds. Have me read the captions against what I actually said, then download the captions file BEFORE pressing the burn-in button, so the words are safe whatever happens next. 3. The check that matters: have me confirm the tool caught our program names and place names, and fix them in the editing pane. If it got those wrong in a thirty-second clip, it will get them wrong in a ten-minute one, so we fix the tool's word list before I go further. 4. Only then, a real video. Remind me it will be slow, that the slowness is the price of the video staying here, and that the editing pane is a step I do every single time. RULES - Never ask me to upload, attach, or paste a video, a frame, or a transcript from one into this chat. Descriptions only. - If this product or my computer genuinely can't run this, say so plainly now and tell me what my options are, rather than handing me something that fails halfway through a render. - If anything fails, explain it in plain words and give me a fix I can do by clicking, or a different route: the captions file uploaded alongside the video, or a free desktop program that burns a captions file into a video. Never send me to a terminal, a local server, a developer console, or into the code, and never use words like SharedArrayBuffer, CORS, or localhost with me. - Before we finish, remind me that once the captions are burned into the picture they can't be edited: the fix is doing the whole render again. So I read them before I press go. - If an organizational setting blocks a step (sharing, permissions, an admin restriction), never suggest a personal account or any other way around it. The only options are asking whoever administers that setting, or a different method entirely. IF THE VIDEO IS PUBLIC ANYWAY AND SPEED MATTERS MORE Tell me plainly that uploading it to a captioning service is a reasonable choice, and help me think it through rather than talking me out of it. This tool is for the videos I would never upload, and for getting off per-minute pricing. Start now by telling me, in two sentences, what we're going to make together, then ask your first question.
Local isn't always the right call, and that's fine. If a video is public anyway and speed matters more than privacy, uploading it to a captioning service is a reasonable choice. The prompt makes the assistant say so instead of quietly building you a slow tool you didn't need. This recipe is for the videos you'd never upload, and for getting off per-minute pricing.
What you walk away with
A downloaded video with captions burned in, and a reusable tool for the next one. No per-minute fee, and nothing uploaded.