I don't make videos. Not one. Not because I don't want to — I've wanted to for years — but because there isn't a single step in the process I can do on my own, and I got tired of finding that out one step at a time.

Start right at the beginning. I point a camera at my own face, and I have no idea whether my head is in the shot.

Not "roughly, probably." No idea. The camera tells you nothing. You'd sit down, talk for eight minutes, send the file to somebody who can see, and they'd tell you the top of your head was cut off the entire time. So you do it again. Maybe that one's fine. Maybe the light behind you turned you into a silhouette and nobody thought to mention it, because to them it was obvious.

That's the easy problem, by the way. The hard one starts once you've got usable footage and have to cut it together, and that's where everyone who has tried this hits the same wall: the software wasn't built with us in mind, and mostly can't be talked into it afterward.

So this week I spent four days building an editor that solves both. It's called Accessible Video Editor, which is not a clever name, and that's on purpose.

What it actually is

It's a video editor where a blind person is the intended user — not the person you write an accommodation for once the layout has already been decided.

Three things make it different from everything else. You cut the video by editing its transcript, like a document. You move around the timeline by ear, with sounds instead of a picture. And you frame the camera by listening to a tone that goes quiet when you're centered.

Everything else in it comes from those three ideas.

What already exists, and where each one stops

One bit of vocabulary first. A screen reader is the program that speaks what's on the screen — it's how I use a computer at all. Mine is Orca, which is the screen reader on Linux; I run Gentoo. On Windows people use NVDA or JAWS, on Mac it's VoiceOver. When I say a program "isn't accessible," I mean one of those can't get you through it.

I want to be fair below, because some of these are good software.

Premiere, DaVinci Resolve, Final Cut Pro. These are the industry. They're also, for practical purposes, unusable by a screen reader user. This isn't a hot take — it's the finding of published research on the problem, which points out that huge parts of the work (layering clips, drawing, most timeline manipulation) simply require a mouse and eyes. Adobe publishes an accessibility page and I don't doubt the effort behind it, but you cannot get a video out the other end. On the Apple side it's not much better — blind users on AppleVis have been reporting for years that the part of iMovie where you cut a clip in the middle isn't reachable with VoiceOver at all. The workaround is to trim your clips somewhere else first, which is a bit like a recipe that starts "have someone else cook it."

Reaper, with OSARA and SWS. This is the real one. This is what blind people actually use, and it deserves respect. OSARA is an extension that makes Reaper — an audio program — genuinely, deeply accessible, and it's some of the best work anyone has done in this space. People do cut video in it. I'll come back to this properly in a minute, because it's the only honest comparison.

Descript. Type-to-edit: it transcribes your video and you delete words to delete footage. That's a genuinely good idea and I've stolen the core of it. But editing text is only one part of making a video, and it does nothing about the camera pointing at your face.

OBS. I should be straight about this one, since I've been describing my own thing as an accessible OBS. OBS is already reasonably accessible. Blind people stream with it. There are rough edges — the filter list doesn't read properly, and getting focus into the sources panel is a fight — but the OBS team has done real work and it mostly holds up.

The research projects and the one-person builds. There's AVscript, an academic prototype from a 2023 conference that does text-based editing with visual problems flagged in the script. Good idea, not a thing you can download. And there's Ultra Creative Suite, built by another blind developer for Windows — a working demo on GitHub, aimed mostly at generating video from audio.

So the honest state of the world is: one excellent workaround, one good idea trapped in a subscription web app, some prototypes, and three industry-standard tools that don't work at all.

The Reaper comparison, honestly

Reaper with OSARA is a beautifully accessible audio editor that will also render video. That sentence is the whole thing.

You can genuinely edit in it. Split clips, move them, reorder them, crop them, copy and paste — all of that works, because to Reaper a video clip is just an audio clip that happens to carry a picture. That's a real insight and it's why the approach works as well as it does.

Where it stops is everywhere the picture actually matters. Reaper's video effects live in something called the video processor, and the accessibility wiki's own video editing guide says it plainly: "it's not 100% accessible," and "it's best to have real time sighted guidance or feedback when tweaking parameters." Putting a title on screen means opening a preset and editing its code by hand to change the font and where the text sits. Then you still don't know what it looks like.

Here's what I can do in mine that I couldn't do there:

• Frame a shot before recording it. Press a key and you hear a tone that pans to where your face is and rises in pitch as it moves up the frame, ticking faster as you get too close. When you're framed, it goes silent. Silence is the target — you stop moving when the sound stops, rather than interpreting a reading.
• Hear my own levels while I talk. Not a number I ask for afterward — a continuous tick whose pitch follows my voice, so the shape of my delivery is audible. It only uses words when I cross into a new zone: "yellow," "red," "clipping." A meter that talks constantly is a meter you turn off.
• Ask what the footage looks like. One key measures a stretch and tells me it's under-exposed, that the highlights are blown, that there's a warm color cast, that the audio floor is noisy, that I peaked at half a decibel and effectively clipped. Another key describes the actual frame — how I'm framed, what's distracting behind me, whether on-screen text is clear of my face.
• Cut by editing the transcript. The video is transcribed word by word. Delete a line, that footage is gone. Move a line up, the video reorders with it. Typing edits the captions and never the cut — a rule the app tells you the first time you type, because it surprises everyone.
• Put a title on screen by pressing a key, rather than editing code.
• Know what an export will do before it does it. Choosing a vertical preset says out loud that the sides get cropped and you'll lose about 68 percent of the frame. That's normally the sort of thing you find out later, from a comment.

What Reaper has that I don't: twenty years of hardening, thousands of daily users, a real audio engine with real plugins, and Windows and Mac support. It's shipping software. Mine is four days old. If you need to cut something this week, use Reaper.

Is it an accessible OBS with Premiere and Photoshop bolted on?

That was my own shorthand for it, and it's about seventy percent right. Let me fix the other thirty.

The pieces are there. There's an editor with a timeline, tracks, splits, transitions, fades, titles, b-roll and rendering — the Premiere part. There's an image editor where you crop, straighten, resize, correct color and read the histogram, and where you draw by describing what you want ("circle at centre, radius 20 percent, white") because painting is a gesture and a sentence isn't — the Photoshop part. And there's a streaming view with scenes, sources, multiple destinations at once, and live chat from Twitch, YouTube and Facebook, each in its own pane so a reply can't go to the wrong audience — the OBS part.

Where the shorthand is wrong is "bolted on." They're not three apps sharing a window. They're one app because they're one problem, and the problem is doing something visual without being able to check it. The same measure-and-say-it-in-plain-words logic that tells me a clip is under-exposed also tells me a scanned photo is sitting 2.4 degrees crooked and filling 38 percent of the scan. The same tone that tells me where my face is in the camera also tells me where a pointer is on a photograph. Building them separately would have meant solving that three times, slightly differently, which is how you end up with an app where every corner behaves like a different app.

The other thing that phrase undersells: the goal was never to make those three programs readable. It was to figure out what each of their controls is actually for, and then offer that instead.

Where it actually is

Version 0.16.0. Sixteen chunks of work, about thirty thousand lines of code, around 800 automated checks that run every time I change anything. Every key that's listed does something — an audit found nine keys documented as working with nothing behind them, and a key that does nothing doesn't read as a missing feature, it reads as your keyboard not working.

It is deliberately not 1.0, and the reason isn't a missing feature. It's that I haven't cut and published a whole video with it yet — which, given how this post started, is rather the whole point. That's the only test that finds the things nobody thought to check for. Until I've taken something from camera to YouTube inside it, the version number stays where it is.

It's also Linux only right now. I tried a cross-platform toolkit first, built the whole front end on it, launched it, and Orca told me there was nothing on the screen — not the window, not even its title. That was two days of work in the bin. A Windows version is planned and I know how I'd do it, but nothing gets built until I've made a four-track scrubber that reads correctly under NVDA. I've paid for that lesson once already.

The thing I'd tell anyone else building this

Every time I tried to translate a visual control directly into speech, it was bad. A meter that reads out numbers is useless while you're talking. A curve editor where you place points by keyboard is technically operable and practically miserable.

Every time I instead asked "what decision is this control actually for," the answer worked. You don't want a volume curve, you want to duck the music under the narration — so the app offers shapes with names and says the number afterward. You don't want the resize dialog, you want it to fit 1080 — so that's the preset, and "1920 by 1080" is arithmetic you shouldn't have to do first. You don't want a level reading, you want to know if you're too loud.

The accessible version of a thing isn't the visual version with labels stuck on. It's the underlying decision, offered directly. Which turns out to be a better interface for everyone — the feature I most want to build next is a properly drawn timeline for sighted collaborators, and it's the easy half, because the accessible model came first.

The code's at github.com/churst90/accessible-video-editor. It's a personal build, so there's no installer and it assumes a few things are already on your machine, but the manual is thorough and the design notes explain why every decision went the way it did. If you're a blind creator who's been told to just use Reaper, or a sighted developer who's been asked to "make the editor accessible" and has no idea where to start, that's the useful part.

And if you know a way to tell whether your head is in frame that doesn't involve asking someone, I'd genuinely like to hear it. I spent four days on my answer.

The other things I build

Same place all of these come from — software I needed and couldn't buy:

• AURA — a free, open source screen reader for Windows, built on a modern foundation. Early days, and honest about it.
• Accessible Trade Terminal — a full trading and charting platform you run by keyboard and ear, with the chart itself turned into sound. Free browser demo, no install.
• Accessible Weather Center — The Weather Channel's old local forecast systems, recreated so that every screen, including the radar, is something you can hear.