Case file 03Things I've built

Vanish

Hide a date stamp, old logo or handle on video you own, privately, in the browser.

Open in the office →The same case file, inside my interactive office

In shortA private, in-browser watermark remover for your own video. I designed and built it, from an overclaiming prototype to a live product at withvanish.com.

Situation: Creators sometimes need to hide a mark on their own footage. Pro editors take skill and a licence, and online removers upload your video. My own first prototype also claimed more than it did.
Task: Vanish is my own product. I did the product, UX and UI, front end, brand and launch, directing AI coding and research agents.
Action: I named every fill for what it really does instead of calling it AI, rebuilt export so it keeps the resolution and sound without leaving the browser, chose a monochrome interface, and opened a waitlist before the app.
Result: It went from audit to a live prelaunch in about two days with 65 tests passing, and it’s now live at withvanish.com with payments switched on.

Client: Vanish (my own product)
Category: Product · in-browser tool
Role: Designer and builder: product, UX/UI, front end, brand and launch
Team: Me, directing AI coding and research agents
Timeline: Prototype from 1 Jul 2026; redesign, rebuild and prelaunch 6–8 Oct 2026
Platform: Web, running entirely in the browser
Tools: HTML/CSS/JS, WebCodecs + Mediabunny, Cloudflare Pages + Functions, Neon Postgres, Resend, Paystack
Status: Live

Live ↗withvanish.com

Creators sometimes need to hide a mark on their own footage: an old @handle after a rebrand, a camcorder date stamp, a retired logo. Pro editors take skill and a licence. Online removers want your video on their server. Vanish does it in the browser. Draw a box over the mark, pick a fill, check the before and after, then export. Nothing uploads, there’s no account and there’s no AI.

Stage 1Situation

01The problem with the prototype

I started from a prototype that looked polished but overclaimed. Its fills were labelled “AI Generative Inpainting” and “FFmpeg style”, but they were simple canvas operations. The “before” view drew a fake logo box. Export recorded the screen at 30 fps with no sound. I audited it line by line and wrote down what it really did against what it claimed.

Stage 2Task: my part

02My part

Vanish is my own product. I designed and built it: the product, the UX and UI, the front end, the brand and the launch. AI coding and research agents did much of the typing and the research under my direction. The decisions below were mine.

Stage 3Action: the decisions

03Decision 1 · Name every fill for what it does

I could build real AI inpainting or rename the fills honestly. I renamed them: Copy nearby, Blend, Pixelate, Solid colour. The privacy promise only works if every other claim is true too. The trade-off: less buzz than “AI”, and the fills work best on static corners.

The editor in Compare view: the outline shows exactly where the fill copies from

04Decision 2 · Real export, still in the browser

The prototype’s screen recording lost resolution and sound, and bundling FFmpeg meant a heavy download. I used WebCodecs instead. The export keeps the resolution and the original sound and never leaves the tab. The trade-off: it needs a current browser, so older ones get a clear message.

05Decision 3 · A calm tool, not a costume

I built six dashboard directions. Five relied on a colour or a theme to be memorable. I picked the sixth, Pure White: monochrome, so the user’s footage is the only colour on screen. The motion and the dashed box carry the brand.

First-visit coach marks: four short steps, replayable from help

06Decision 4 · Keep the promise in the business model too

Pro keys are signed (Ed25519) and checked in the browser. There’s no account and no user database, so “nothing sent from this page” stays true. The trade-off: a check in the browser can be bypassed. At ₦2,000 a month, it’s built for honest users.

07Decision 5 · Open the waitlist before the app

Prelaunch shipped only the landing page and a waitlist, and the build fails if an app link or payment URL slips in. The waitlist prints a ticket with your position and whether you got one of the first 20 free Pro months.

The printer-ticket waitlist (test address)

404: the mistyped address is treated as the mark. Drag the line and it’s gone.

Stage 4Result

0 BVideo uploaded: the whole edit runs in the tab
~2 daysFrom auditing the prototype to a live prelaunch (6–7 Oct 2026)
65Automated tests passing at prelaunch

08Result

Vanish went from an audit of the prototype to a live prelaunch in about two days, with 65 automated tests passing. It’s now live at withvanish.com with payments switched on: a paid product, designed and shipped by one person.

What I learned

Honesty is a design constraint, and a good one. Dropping “AI” forced clearer labels, a better compare view and a stronger privacy story.
One device beats many. The dashed box gave the product, the brand and the marketing one idea to repeat.
Building three to five working variations side by side settles taste questions faster than mock-ups.
AI agents speed up building, not judgement. Picking the right footer, and rejecting the ones that looked like someone else’s, was still my job.

Footage by Pexels creators, credited shot by shot in the project. Fonts: Geist and Geist Mono by Vercel (OFL). Export muxing: Mediabunny (MPL-2.0).

SStage 1Situation01 / 04