Score Your Reel Before You Post

How I run the Higgsfield Virality Predictor from Claude Code, and what 18 scored versions of my own reels taught me about hooks.

0101

What it measures

You upload up to 15 seconds of video and it predicts how a human brain responds, second by second. You get an overall viral score out of 100, a hook score for the first 3 seconds, a retention (sustain) score, and a per-second response curve that shows where attention dips. On Oct 10 2026 it cost 0 credits on my Higgsfield starter plan. Treat it as a pre-screen that narrows your variants, not a promise that anything goes viral.

0202

Run it from Claude Code

Connect Higgsfield to Claude Code through its MCP server, then let Claude do the busywork. Claude trims a scoring copy to 15 seconds with ffmpeg, uploads it, runs the predictor (the job shows as brain_activity, about 2 minutes) and reads the scores back to you. Paste the prompt below into Claude Code and point it at your reel file. Ask it to log every result to a scores file so you can compare versions later.

0303

Test hooks, not reels

Change only the first 3 seconds, keep everything else identical, and score 3 versions. The model is deterministic: the same clip gets the same score every time, so any difference comes from your edit. It also scores each clip against its own range, so a hook number only means something next to another version of the same reel. Test the opening line, the first visual and whether text sits on screen in second one.

0404

What 18 versions showed

I scored 18 versions of my own reels. The spread was 46 to 56. Putting text over my face in the first second dropped the hook from 38 to 27. My real face, steady, one voice, scored highest (56). Split screen landed at 50, jump-cut punch-ins at 52, and slow push-ins changed nothing. Shorter clips (9 to 11 seconds) held retention at 100. The topic barely moved the score.

0505

Use the score to win clients

Run a brand's own ads through it and put the curve in your pitch. Score their current ad, make one sample with a stronger hook, score that too, and send both numbers with your sample link. A brand sees exactly where people drop off in their ad and what your version fixes. One warning from experience: a reel edited only to please the score turned bland. Build for craft first (big captions, a new visual every 1 to 1.5 seconds, proof full size), then use the score as a final check.

Copy · paste · send

Score my reel with the Higgsfield
Virality Predictor.

1. Make a 15 second copy of
[reel.mp4] with ffmpeg.
2. Upload it to Higgsfield and run
virality_predictor on it.
3. Give me: overall, hook,
sustain, and the second where
attention dips most.
4. Append the result with a label
to scores.jsonl.
5. Suggest one change to the
first 3 seconds, then stop.

Want brands to pitch?

Client Finder finds them in your niche, with their emails. It is in the free trial.

Find your client →

Not ready for that?

Drop your email and 3 more guides open in the Library, free.

Unlock all 84 guides

Find your client →