The Human Attention Model.

The retroactive trap.

Every video published today is judged twice: once by a recommendation algorithm and once by a human nervous system. Both judgements happen after publish — long after the only window in which a creative director could have prevented a $50k shoot from flopping.

The current state of practice is post-hoc analytics: ship the video, watch the retention curve, and guess at what to do next time. By the time a strategy team has enough data to learn the lesson, the production budget has already been spent. The result is structural: an industry that guesses blindly, wastes days of editing time, and optimises for the algorithm because the algorithm is the only thing it can measure.

The Old Way vs. Soma.

Creative teams have historically relied on subjective feedback and retrospective data to inform future cuts. Soma replaces guesswork with a deterministic simulation.

Three things the brain reliably does.

Soma rests on three findings from cognitive neuroscience that hold across language, genre, and culture.

1. Bounded attention windows. Working memory tolerates ~3–7 seconds of unstructured stimulus before disengagement. Most "first-second" hooks miss because they violate this window, not because they're boring.
2. Pattern interrupts compound. Disengagement is not gradual — it's punctuated. Each interrupt either resets the window or drops the viewer through it. Density and timing matter more than novelty.
3. Cognitive load is two-tailed. Both overwhelm (too dense) and boredom (too sparse) trigger drop-off. There is a measurable "optimal load" zone, narrow but stable across viewers.

The Grounding Engine.

Soma is not a generic LLM wrapper. It does not guess retention based on text prediction. The 500M+ persona simulation is deterministically mapped to the true behavior of global audiences.

Grounded with real human behavior and conversations. For every frame, the engine correlates the on-screen pixels and audio prosody with the exact second viewers dropped off, and the exact timestamps where they commented. When you run a cut through Soma, you are simulating the precise, statistically validated reactions of half a billion viewers — what they do (watch-time, scroll depth, skips) and what they say (comments, sentiment, shares).

Soma in four stages.

A video enters Soma as three parallel modalities — pixels, audio, transcript — which pass through a multimodal encoder into a massive parallel simulation of 500M+ grounded audience personas. The simulation exits as five attention signals plus a recombined retention curve.

Each modality is first encoded into a shared attention space. This fused representation then drives an agentic simulation of 500M+ audience personas. These are not generic LLM agents, but hyper-specific audience replicas grounded by trillions of real-world data points — capturing precisely what they do (watch-time, clicks, scroll depth) and what they say (comments, sentiment, shares). Frame-level outputs from this simulated audience are recombined into a single retention curve, accurately forecasting viewer behavior.

The retention curve, returned millisecond-resolved.

Soma's headline output is a per-frame retention forecast. By predicting the exact shape of the curve before export, strategy teams can salvage dropping retention and avoid burning thousands of dollars on ineffective media spend or dead video concepts.

The curve is segmented into bands: the 5-second cliff (where ~40% of all uploads lose >50% of viewers), the mid-video slump (the 18–28s zone where pacing fatigue compounds), and the recovery shoulders where pattern interrupts can rescue an otherwise descending trajectory. Soma flags these structural leaks and suggests millisecond-precise cuts, turning what used to be a week of trial-and-error editing into a single targeted revision.

Eight signal categories.

Each Soma inference produces eight per-frame signal channels. By separating these signals, creative directors don't just know where viewers drop off — they know exactly why, allowing for precise fixes rather than blindly reshooting.

The two-tailed problem.

Cognitive load is the most misunderstood signal in video. Editors instinctively maximise stimulus density; viewers leave when stimulus exceeds working-memory capacity. Soma scores this on a normalised 0–100 scale per frame, with five interpretable bands.

Editors' single most common pattern of failure is parking entire videos in the high-load band. Soma surfaces these stretches as red bars on the timeline, with cut suggestions to redistribute load — usually by extending pauses, not by removing content.

Soma Lite vs Soma Pro.

Two model tiers ship today: a fast, single-channel cognitive scorer (Lite) and a full multimodal architecture (Pro).

What we measured.

Soma is benchmarked against held-out human viewer panels on a 12,000-video corpus across six verticals, plus continuous shadow-evaluation on creator opt-in production data.

What Soma does not do.

Soma forecasts attention; it does not forecast intent. A high-retention video may still under-convert, and a polished Soma score does not guarantee distribution under any specific platform's recommendation policy. Soma does not replace editorial judgement on taste, brand fit, or legal review. It also does not predict virality in the strict sense — virality is a network-propagation phenomenon that depends on factors outside the video itself (timing, social graph, distribution).

Soma's signal degrades on extreme out-of-distribution content (non-narrative interfaces, generative-only abstract video, screen recordings without spoken language). For these surfaces, Pro returns confidence-weighted scores instead of point predictions.