Predict a reel's performance before you post it, with Claude Code

A reel is won or lost in three seconds. Instagram measures exactly that: the share of people who swipe away from your video before the third second. The problem is that you find out after posting, when the shoot and the edit are already done.

This guide flips the order. You test your hooks before you publish, with a model that estimates an audience's attention second by second, right from Claude Code. Then you confirm with Instagram's real data.

What the predictor measures

Higgsfield's Virality Predictor does not guess a view count. It has a modeled audience watch your clip and maps its brain response: vision, sound, memory, attention, language. You get an estimate of the attention your video holds, second by second.

That is the signal you need to pick between three hooks: comparative, fast, and available before your audience ever sees the video.

Don't want to cut by hand? Ask Claude Code, it uses FFmpeg (installed on your machine):

Cut the first 15 seconds of version-a.mp4, version-b.mp4 and version-c.mp4 with ffmpeg, and save them as hook-a.mp4, hook-b.mp4 and hook-c.mp4.

Analyze these three clips with Higgsfield's Virality Predictor: hook-a.mp4, hook-b.mp4, hook-c.mp4.
For each one, give me the overall score, the peak hook and its second, the hold, and the weakest region.
Rank the three hooks from strongest to weakest.
For the best one, tell me what happens on screen and in the text at the second where attention drops, then suggest three specific changes for its first 3 seconds.
Present the scores as attention estimates, never as a view count.
Give me the report link for each clip.

Under the hood, each analysis is this command, which you can also run yourself:

higgsfield generate create brain_activity --video ./hook-a.mp4 --wait

An attention model is still an estimate. The log tells you how far to trust it on your own audience:

Create a predictions.csv file with these columns: date, reel, hook, overall score, peak hook (second), hold, real skip rate, views at 24 h.
Add one row for each clip you just analyzed, with its Higgsfield scores.
When I give you a reel's Instagram insights, fill in its last two columns.
After 10 reels, tell me whether the predictor's ranking matches the real ranking by skip rate.

What this changes for your business

Every reel costs a shoot, an edit and a slot in your calendar. Testing the hook before you publish means you stop paying those hours for a video people swipe past in two seconds.