TECHNOLOGY
A brain-encoding model.
Running at commercial speed.
Endurance runs its proprietary neural model — a trimodal brain-encoding architecture trained on fMRI, EEG, and eye-tracking data — to predict how any creative asset lands in the human brain before it reaches a single audience member.
average time per creative simulation
brain signals modeled per frame
prediction accuracy vs. post-campaign recall data
modalities — attention, emotion, memory-desire
THE PROCESS
Three steps from creative brief to brain map.
Every Gargantua simulation runs the same deterministic pipeline. You ingest a creative. We decode how a human brain processes it. You get an actionable output.
Upload your video, still, or copy variant via the Labs workspace or the Gargantua API. The pipeline accepts MP4/MOV (up to 4K), PNG/JPEG/WEBP, and plain-text copy blocks. No file prep required.
The frame sequence passes through the Endurance neural encoder — a transformer-based trimodal encoder fine-tuned on a proprietary fMRI + EEG + eye-tracking corpus. The model predicts the neural response across three dimensions for every frame or semantic unit.
Tensors are decoded into four human-readable outputs: a spatial attention heatmap overlay, a frame-by-frame emotional valence curve, a memory encoding score, and a reward / desire signal. Delivered as a JSON payload and as an interactive visual report in Labs.
THE MODEL
The Endurance neural model — trimodal brain-encoding
The Endurance neural model maps visual input directly to predicted fMRI activation patterns. We extended a benchmark-grade brain-encoding architecture with proprietary fine-tuning data across video, copy, and emotional labeling.
"Brain encoding models represent the most direct line from stimulus to predicted human response ever deployed outside a research lab."
— [SOURCE / CITATION]
level response
THE SCIENCE
Three dimensions of cognitive response.
Every creative asset activates attention, triggers emotion, and encodes memory at a different rate. Gargantua models all three — not as a composite score, but as independent channels you can optimize against.
Predicts where the eye lands, how long it stays, and what it skips — frame by frame. Output is a spatial heatmap overlaid on the original creative. High-attention zones are color-coded from cold (ignored) to hot (fixated).
Measures predicted emotional valence (positive/negative) and arousal intensity across the temporal span of the creative. Delivers a per-second curve — you see exactly where a 30-second spot peaks, drops, or flatlines emotionally.
Encodes how likely any element — brand mark, product shot, headline, CTA — is to consolidate into long-term memory, and whether it activates the reward / desire circuitry associated with purchase intent.
USE CASES
What Gargantua tells you — before you spend.
VIDEO ADS
Is the brand landing in the first 5 seconds?
The average viewer decision on a pre-roll happens at second 4. Gargantua tells you whether the logo, product, or key message hits the attentional peak in the open window — or whether you're burning creative on a brand moment that happens at second 18.
DISPLAY / OOH
Static creatives — tested before production spend.
Upload 5 banner variants and know which one drives the highest brand mark fixation and purchase-intent signal before allocating a media line item. OOH format support: billboard, transit, DOOH.
PACKAGING
Shelf attention before the tooling invoice.
Simulate how a packaging redesign performs in a shelf context — competing against known brands. Know if the new logo placement gets noticed, or if the product name is lost in periphery, before the design goes to print.
CONTENT MARKETING
Long-form content — tested for cognitive load.
For editorial campaigns, whitepapers, and branded content, Gargantua measures how memory consolidation accumulates across the reading arc — and where attention drops. Identify the exact paragraph where the reader mentally leaves before you publish.
WHY NOW
The compute barrier collapsed in 2024.
Brain-encoding models existed in neuroscience labs for a decade. Running them at commercial speed and cost was impossible before three things converged.
01
Open-weight large vision models
Brain-encoding transformers trained directly on fMRI data at scale have crossed from research benchmark to deployable prediction — now runnable on consumer-grade hardware with quantisation.
Inference cost fell below $1 / creative
H100 spot pricing + 8-bit quantisation brings a full Gargantua simulation run — frame-by-frame attention, valence, memory, reward — under $1 per creative. Two years ago the same workload cost $40–60 on-demand and took 30+ minutes.
Proprietary fine-tuning corpus
Endurance has assembled a fine-tuning corpus of [1k] annotated campaign assets across [X] categories, paired with post-campaign recall and conversion data. This is the moat — the base model is open, but the fine-tuning data is ours.
GET STARTED
See your campaign's mind.
Request a live demo of Gargantua on your own creative assets — no prep required.