FourFaced

One clip. Four faces. Every caption backed by what's actually on screen.

REFRESH

MANUAL

DARK

Drop a clip here, or choose a file

MP4, MOV, WebM

No clip handy? Try one of these samples.

Pick the Four Faces you want captioning your clip

MODEL IN USE

Formal

Polished and professional — ready for a press page.

Sarcastic

Dry wit with a raised eyebrow.

Humorous — Tech

Jokes only your dev team will get.

Humorous — Everyday

Casual laughs for any feed.

Four Voices

Evidence — what Gemma actually saw

FourFaced

Team VERPTO — AMD Developer Hackathon ACT II, Track 2

USER MANUAL

FourFaced

One clip in. Four captions out.
Here's everything you need to know.

Click the page (or press Next) to open →

What is FourFaced?

You give it a short video. It watches the video, then writes four captions for it — each one in a different voice. We call them the Four Faces:

01 FormalPolished and professional

02 SarcasticDry wit, raised eyebrow

03 Tech humorJokes for computer people

04 EverydayCasual laughs for anyone

Pick the caption you like best, copy it, and post it anywhere.

01 1

Step 1 · Add your clip

🎬Your clip drag → 📥Drop box → 📺Preview

Drag a video file into the dashed box — or click the box and pick a file from your computer.

No clip handy? Click one of the sample pills below the box.

You can add up to 4 clips — they line up like characters on a select screen. Click one to bring it to the front:

clip 1 clip 2 clip 3 clip 4

02 2

Step 2 · Caption it

✓ FORMAL ✓ SARCASTIC ✓ TECH ✓ EVERYDAY

All Four Faces are ticked to start. Untick any you don't want — its card below goes dim so you can see who's sitting out.

✅Tick Faces → 💬Caption → ⌛~30 sec

Then press the big Caption button. Each clip takes about half a minute; with several clips they go one after another while you watch the progress line:

03 3

Step 3 · Use your captions

Each Face gets its own card, like this:

FORMAL

A sunlit street lined with autumn trees…

CopyEdit🔊

Copy puts the caption on your clipboard. Edit lets you change any words (Save keeps them). The speaker icon reads the caption out loud, each Face with its own voice.

The Evidence box underneath shows what the AI actually saw in your video — so you know the captions aren't made up.

04 4

Handy tips

The × on a video removes that clip.
With several clips, use the ← → arrows, the dots, or click a side clip to switch. The captions shown always belong to the clip in front.
Drag more videos anywhere onto the card to add them — no need to hunt for a button.
If a clip says FAILED, the others still finish. Just press Caption again.
Captions describe what's really on screen — an odd clip gets an odd caption. That's honesty!

05 5

The models & the backup plan

Every caption starts with Gemma 4 31B — a big AI model that can see. It watches your clip, notes what's really there, and writes all four captions.

🧠Gemma 4 31B keeps failing? → 🛟Gemma 4 26B

If 31B keeps erroring (with time left in the ~30-second promise), Gemma 4 26B gets one try — at any step: watching, writing, or the final double-check.

Both couldn't watch the clip?You get a clearly-marked stand-in caption — it never pretends it saw something.

Watched it, couldn't add the voices?Captions are built from the real notes it already took — plainer, but still true to your video.

The Evidence box always shows which path — and which model — did the work.

06 6

That's the whole book

Close the manual and go caption something. If a result ever looks off, the Evidence box will tell you exactly what happened — including which model did the work.

Made by Team VERPTO
AMD Developer Hackathon ACT II

Cover