proof
Receipts, not assertions.
Anyone can say “broadcast-ready.” Here is exactly what gets measured on every spot, what verified it, and — just as honestly — what we haven't built or turned on yet. Nothing below is rounded up.
≈4 min
target: draft spot, URL in
55
automated checks that can block a spot from shipping
900
frames in every :30, exactly
1555
agent tools — the whole platform
286 (sample, 2026-10-10)
capabilities shipped
The draft-spot time is our engineering target, not an average — real runs vary with scene count, tier and revisions. Every other figure is read from the platform’s own registries, or labelled a sample.
Your own picture
The sketch drawn
480 draft: coming
The 480 draft
1080 finish: coming
The 1080 finish
03 480 draft: coming · 04 1080 finish: coming
live today
Independent model families review every scene. None sees another's answer.
A same-family ensemble measures nothing — correlated models fail together. Every scene is judged by models from different vendor families, run in parallel, blind to each other, on one shared rubric.
what the ensemble checks
- On-screen text and marks — ordinary in-scene lettering is checked for being legible, correctly spelled, and belonging in the scene; your own brand's text and marks are checked against your cited facts and real reference art. Anything wrong, or anything nobody can confirm, holds for a person to check — never a silent ship. Logos, phone numbers, prices, and offers are never AI-generated pixels at all — they're composited real assets.
- Motion plausibility — physics, continuity, and direction measured across the sequence, not guessed from a single still.
- Product and brand accuracy against your own brief.
- Interpenetration and occlusion — a body walking through furniture, a figure passing through a doorframe.
A scene that fails retries with increasingly specific direction — and a stronger model if it must — up to a bounded limit, never an unlimited loop. Anything the ladder still can’t resolve holds for a person to review — it is never shipped silently.
the validation report
- Frame-exact duration — a :30 is measured in frames, not seconds (900 frames at 29.97, not “close enough”).
- Loudness normalized to −24 LKFS ±2 and re-measured on the finished file, not just applied in the mix.
- Your station’s own spec instead, when it differs — set the delivery target once (EBU R128’s −23 LUFS, say) and every master lands on it and is measured against it.
- CEA-608/708 captions embedded in the broadcast file, plus SCC/SRT/VTT sidecars.
- Safe areas, real broadcast codecs, and XDCAM/ProRes conformance on the frame-exact masters.
live today
A delivered spot never ships with a failed validation report.
Every render carries a validation report — duration, codec, loudness, and captions all checked before the file is ever exposed. A render whose report failed is never handed to you; you see the honest flag instead.
live today
C2PA content credentials, signed on delivery.
Every delivered MP4/MOV/MP3/WAV master is signed with a C2PA manifest naming the AI-generated elements and the transcode chain — a durable, tamper-evident record of how the file was made.
Said plainly, not oversold: our signing chain is self-anchored today — a third-party verifier that checks the credential will show “unknown source” until we’re issued by a public trust-list authority. The signature itself is real and tamper-evident now; broad third-party trust is a separate step we haven’t taken yet, and we say so rather than imply it.
engine shipped · ordering coming
Print and display ads get a measured press-conformance report.
The same engine that makes your TV spot builds display statics and press-ready print ads — sized, arranged, and separated for CMYK print — with a conformance report that states exactly what was measured.
measured and passed
- Every font embedded; ad copy is live vector text, not a flattened image.
- CMYK/grayscale only — no RGB colour survives to the press file.
- Total ink coverage under the publisher's limit, and black built from the K plate — never a four-plate rich-black that won't dry on press.
- Page count, page size, and effective image resolution against the publisher's own floor.
11 of 13 conformance checks are measured and reported on every print-tier render.
named, not claimed
- No PDF/X preflight runs — we never print a PDF/X-1a or PDF/X-4 badge. The report states the flavor the publisher asked for beside exactly what we measured.
- A licensed press ICC output intent (e.g. SNAP2007) isn't vendored — that check reports “not run,” not a guess.
- Quick Spot orders web and social display ads from one website address — one ad, or three headline ideas from the same read to pick between; print sizes are ordered in the studio's print room, not from Quick Spot.
the engines
One frame, and every engine behind it. Named, not hidden.
A shot doesn't go to one AI vendor and hope. The router picks an engine per shot, independent reviewers from different vendor families judge it, and any compromise is disclosed on your own timeline — never a silent swap.
coming A frame from the showcase spot lands here once it is exported.
video generation
- Veo 3.0/3.1 (Google Vertex AI / Gemini Developer API)measured challenger
- Runwayrouted
- Klingrouted
- Seedance 2.5 (BytePlus ModelArk)routed
- PixVerserouted
fidelity screening
- Google Gemini (Vertex)live
- Anthropic Claudelive
voice & music
- ElevenLabslive
- ACE-Stepcoming
No engine is load-bearing: 22 routes reach 11 generation models. A route that fails is demoted on its first failure, and the same request moves to another route into the same model.