Evidence-grounded video intelligence

One video.
One truth.
Four voices.

TruthTone understands what happened, and when, before it writes a word. Tone can change. Truth cannot.

4distinct tones
3xcandidates per tone
1temporal truth spine

+Your video is sampled locally; only selected frames are analyzed.

Accuracy engine / built on AMD

Humor gets freedom.
Facts get guardrails.

A three-stage Fireworks AI pipeline separates perception from personality, then challenges every caption before it reaches the screen.

01
Observe

Scene-aware sampling

Uniform coverage plus visual-change detection preserves the beginning, transitions, and outcome.

02
Understand

Temporal Truth Spine

AMD-hosted vision inference records subjects, actions, visible text, confidence, and event order.

03
Compete

Candidate workshop

Three captions per style compete on factuality, sequence, tone, coverage, and clarity.

04
Verify

Critic + repair

Unsupported claims are rejected. Only the strongest grounded caption survives for each voice.