What it does
A pass or reject verdict on every generated video or image ad, from a full watch at one frame per second against ten gates, with the frame numbers as evidence.
Who uses it
What you need
You bring:
- the generated ad
- the script or the claims it makes
- the proven-claims bank for numbers and prices
- ffmpeg
How it works5
The steps, in order. The method itself is served whole through the connector, current on every use.
- 1
Write the claim-to-proof table before generating
- 2
Strip the whole video at one frame per second
- 3
Read every caption on every frame
- 4
Score the ten gates with frame evidence
- 5
Record the table beside the asset
What it doesn't do
- Spot-checking three frames
- Accepting an asset without a gate table
When your AI reaches for it
Quality gate for AI-generated video and image ads (any generator). Use EVERY time a generated ad is about to be accepted, shown to a stakeholder, or queued for upload — and BEFORE writing generation prompts for a new batch. Covers the pre-generation shot contract (every claim must have a visible proof beat), the full-watch protocol (1 fps frame strips of the ENTIRE video — never spot-check), the 10 accept/reject gates, and the proven-claims check (numbers and price framing must come from copy that actually ran and won).