A tool, 2026 — companion to Color fingerprint + Sound fingerprint
Spot Scorer.
Paste any Czech TV spot URL. A multimodal AI reads the picture and the audio in parallel and scores the spot against the 13-brand, 47-spot sector benchmarks. Plus three to five numerical targets the spot would need to hit to read as distinctive.
Data & cloud engineer at Alma Career. Built across a few weekends to test what brand competitive analysis looks like when it starts from public data instead of pitch decks. Part of the five-piece Brand Reflection Series; the only one with a live multimodal AI backend.
Score one video
Paste a YouTube link (most reliable), Facebook public post, or Instagram public post. Pick the brand's nearest sector for benchmark comparison. First call cold-starts the backend (~30s); subsequent calls take 10-25s.
Or compare two spots side-by-side
Score two videos head-to-head. Useful for "us vs them" pitches: paste a client's draft + a competitor's spot, get a paired scorecard showing where they differ on the same sector benchmark.
Try it:
Or generate a brief for a whole category
No URL handy? Pick a category. Get the numerical spec a distinctive new spot in that category would need to hit, derived from where the existing 13 brands actually live. Read this as a starter creative brief, not a verdict.
Color score
Audio score
Sector context
Each gauge shows where this spot sits inside its sector. The shaded band covers the middle half of sector spots; the dashed line is the sector median. A dot near the edge of the band means the spot is already pushing toward distinctive; a dot in the middle means it lives at the category default.
Brief targets
If the goal is a distinctive spot in this category, these are the numerical moves. Each target pushes the spot OUT of the typical sector range on a single axis. Pick one or two; trying to move all of them at once defeats the focus.
What this is and isn't
- Built from 47 Czech TV spots across 13 brands. Sectors with only 3–4 spots have wider, noisier benchmarks — the AI gives you direction in those cases, not precision. If you commission scoring for your client's category, we agree the spot set on the call and tighten the benchmark from there.
- YouTube blocks the scorer sometimes. Less-popular videos and newer uploads can hit YouTube's bot challenge. Facebook and Instagram public-post URLs are more reliable. If a fetch fails, it's an access issue — not a verdict on the spot.
- The AI measures what the spot does mechanically, not whether it's good. Vision AI reads color and frame structure; audio AI reads music vs voice, tempo, key. They catch "the brand color is essentially absent" or "this spot is voice-led when the brief was image-led." They don't grade story, performance, or cultural fit. Use the tool to catch obvious drift in pre-mix, not to grade craft.
- The brief generator gives you a starting spec, not a verdict. The 3–5 targets push the spot out of the typical range on one axis at a time. Hitting them means the spot is mechanically more distinctive in those measurements; whether that's the right strategic move for the brand is the strategist's call.
- One operator, free-tier backend. The AI server is on a personal Modal account. Cold start is about 30 seconds; concurrent users may queue. If you want this in a production workflow with reliability guarantees, that's a deployment conversation, not a public tool.