Beta launching soon

Detect off-brand, slop, low quality content before it embarrasses you.

Brand Grammar Extraction. Reads camera movement, color, lighting, and pacing from your footage into a reusable brand profile.
Quality & prompt adherence. Scores every cut for craft and for whether it delivered what the brief asked.
Brand-kit adherence. Grades any video against your brand kit and flags the off-brand frames, timestamped.

BVI

slop×12

off-brand×8

off-promptdidn't follow prompt×6

qualitygood quality×27

on-promptprompt adherence×19

brand pass×21

sources in · verdicts out — every video routed before it ships

Tier 01 / Capabilities

Capability 01

Brand extraction — the grammar engine

Point it at your videos; it learns the brand's grammar.

sends: POST /brands → refs → extract
gets: frozen brand_profile (versioned)

www.example.com

hero_film.mp4

product_02.mp4

launch_spot.mp4

BVIgrammar engine

extracted brand grammar

camera movementdolly · pan · handheld

color roles#c8503c text · #5b3a21 bg

gradesat 0.41 ±0.06

pacingmeasured, building

editing rhythm3.1 cuts / 10s

compositionclose-up 0.48 · shallow DoF

lightingsoft key, controlled

shot sizesCU · MS · WS

brand markcorner_br · ≥1.5s

typographysans, lower third

tolerancescolor w1.5 · warn 78

style promptwarm earthy palette, shallow DoF, calm camera…

negative promptno brand red in props, no whip pans, no serif body text…

camera movementdolly · pan · handheld

color roles#c8503c text · #5b3a21 bg

gradesat 0.41 ±0.06

pacingmeasured, building

editing rhythm3.1 cuts / 10s

compositionclose-up 0.48 · shallow DoF

lightingsoft key, controlled

shot sizesCU · MS · WS

brand markcorner_br · ≥1.5s

typographysans, lower third

tolerancescolor w1.5 · warn 78

style promptwarm earthy palette, shallow DoF, calm camera…

negative promptno brand red in props, no whip pans, no serif body text…

reference in · brand profile out

GET /grammar/v1/brands/b1…001/profile?version=latest

$ curl -X GET https://api.cognatelabs.io/grammar/v1/brands/b1…001/profile?version=latest \
    -H "Authorization: Bearer $CGN_KEY"

→ 200 OK
{
  "meta": { "profile_id": "…", "extractor_version": "1.3.0",
            "reference_videos": [ { "video_id": "…", "duration_s": 24, "scene_count": 9 } ] },
  "profile": {
    "color":  { "dominant_colors": [
                  { "hex_approx": "#c8503c", "prevalence": 0.12,
                    "roles": ["text"], "is_brand_kit_color": true } ],
                "grade": { "saturation": { "mean": 0.41, "std": 0.06, "p10": 0.33, "p90": 0.5 } } },
    "motion": { "energy": { "mean": 0.38, "p10": 0.2, "p90": 0.55 },
                "energy_bucket_mix": { "calm": 0.6, "medium": 0.3, "kinetic": 0.1 } },
    "editing": { "cuts_per_10s": { "mean": 3.1 }, "transitions": [ { "type": "cut", "frequency": 0.9 } ] },
    "composition": { "closeup_ratio": 0.48, "depth_of_field": "shallow" },
    "brand_mark": { "appears": true, "placement_zone": "corner_br", "min_on_screen_s": 1.5 },
    "qualitative_rules": [
      { "rule": "Brand mark appears in the closing card, lower-right, ≥1.5s.",
        "dimension": "brand_mark", "severity": "hard", "confidence": 0.97 } ]
  },
  "tolerances": { "color": { "weight": 1.5, "warn_threshold": 78, "fail_threshold": 65 } },
  "generation_conditioning": {
    "style_prompt_fragment": "warm earthy palette, shallow DoF, calm camera…",
    "negative_prompt_fragment": "no brand red in props, no whip pans, no serif body text…" }
}

Capability 02

Quality & prompt adherence

Every cut, graded before it ships.

Post a video and its prompt — no profile needed. It scores the craft deterministically (subject and background consistency, temporal flicker, imaging quality, motion bucket) and judges prompt adherence — objects, actions, scene, spatial, text, artifacts — each a { score, status, diagnosis } with pass / warn / fail.

sends: POST /evaluations (video + prompt)
gets: quality + prompt_adherence

Video still being quality-scored — specimen · scorecard UI, reproduced

GET /evaluator/v1/evaluations/e3…333

$ curl -X GET https://api.cognatelabs.io/evaluator/v1/evaluations/e3…333 \
    -H "Authorization: Bearer $CGN_KEY"

→ 200 OK
{
  "quality": {                                         // deterministic scores
    "subject_consistency":    { "score": 88, "status": "pass" },
    "background_consistency": { "score": 72, "status": "warn" },
    "temporal_flicker":       { "score": 41, "status": "fail" },
    "imaging_quality":        { "score": 58, "status": "warn", "p10": 39, "min": 22 },
    "motion_bucket": "medium"                           // calm | medium | kinetic
  },
  "prompt_adherence": {                                 // model-judged
    "overall":  { "score": 64, "status": "warn" },
    "objects":  { "score": 82, "status": "pass", "diagnosis": "surfboard + van both present" },
    "actions":  { "score": 70, "status": "warn", "diagnosis": "van drives but doesn't clearly turn" },
    "spatial":  { "score": 66, "status": "warn", "diagnosis": "van centered, not entering from left" },
    "garbled_text_artifacts": { "score": 55, "status": "warn" },
    "gross_motion_artifacts": { "score": 45, "status": "fail", "diagnosis": "limbs warp; wheel detaches" }
  }
}

Capability 03

Brand adherence — the verdict

Does the cut obey the brand kit?

Supply a profile_id (or an inline profile) on the same evaluation and brand scoring turns on: color, grade, motion energy, editing pace, composition, typography, and brand-mark — each scored against the learned profile with its own distance metric, plus rule violations by timestamp. The always-present verdict carries a badge and machine-injectable recommended_corrections.

sends: POST /evaluations + profile_id
gets: brand block + verdict + corrections

Under review

Verdict

off_brand

hard failure: brand mark missing from the closing card

Brand

Colorwarn68

Motion energypass88

Editing pacepass83

Typographyfail60

Brand markfail30

specimen · scorecard UI, reproduced

GET /evaluator/v1/evaluations/e7…777

$ curl -X GET https://api.cognatelabs.io/evaluator/v1/evaluations/e7…777 \
    -H "Authorization: Bearer $CGN_KEY"

→ 200 OK
{
  "meta": { "profile_id": "11…111", "profile_version": 3 },
  "brand": {                                            // null without a profile
    "color":        { "score": 68, "status": "warn", "distance_metric": "wasserstein_lab_marginals" },
    "grade":        { "score": 74, "status": "warn", "distance_metric": "grade_fingerprint_l2" },
    "motion_energy":{ "score": 88, "status": "pass", "distance_metric": "energy_distribution_overlap" },
    "editing_pace": { "score": 83, "status": "pass", "distance_metric": "shot_length_wasserstein" },
    "typography":   { "score": 60, "status": "fail", "distance_metric": "vlm_typography_rubric" },
    "brand_mark":   { "score": 30, "status": "error", "distance_metric": "vlm_brand_mark_rubric" },
    "rule_violations": [
      { "rule_id": "rule_brand_mark_closing", "severity": "hard", "timestamp_s": 29.0,
        "diagnosis": "No brand mark in the closing card (required lower-right, ≥1.5s)." } ]
  },
  "verdict": {                                          // ALWAYS present
    "overall_score": 58,
    "badge": "off_brand",                               // on_brand | review | off_brand | quality_fail
    "hard_failures": ["rule_brand_mark_closing"],
    "recommended_corrections": [
      { "dimension": "brand_mark",
        "prompt_delta": "add a closing card in the final 2s with the brand mark lower-right ≥1.5s" },
      { "dimension": "typography",
        "prompt_delta": "render captions sans-serif, lower third, <3s; no serif or full-frame text" } ]
  }
}

Tier 02 / Beta

Sign up for the waitlist.

We’re onboarding teams in waves. Tell us who you are and roughly how much video you run — a few fields, no demo call to book.