Fit Check → Conversion Intelligence

The check that runs before your credits do.

A failed render already costs nothing — that's handled. The expensive case is the render that succeeds technically and still underperforms, because the actor was wrong for the product. That one bills the full ~2,000 credits for a finished 15s ad. Fit Check scores the pairing first, from the actual pixels of your product image, and gets sharper every time someone logs what really happened.

💾 Note: logged outcomes are saved locally in this browser (localStorage) — not shared globally, so this demo is safe to run live without another viewer's session mixing in.
~2,000
credits · finished 15s 1080p ad
60–90
credits · per 5s clip
~12
credits · per image
$0
failed renders are free

That last one is the important nuance: the credit refund covers technical failure. A render that completes cleanly but pairs the wrong actor with the product is a success by the system's definition — and still bills in full. Fit Check aims squarely at that gap.

Toolset

Where this would live

Fit Check isn't a separate product — it's a pre-flight step in front of the tools that already spend credits.

URL → Ad
Point it at a product page, get a finished ad back.
Live
Make a custom video
Brief-driven generation with full creative control.
Live
Talking actors
Scripted delivery from an AI, twin, or licensed actor.
Live
Creative Studio
Assemble, edit, and finish the render.
Live
Fit Check
Score the actor-product pairing before you spend credits.
New
Try it below →
The tool itself

Run it — this part is real

Everything below actually computes. Your image is read pixel by pixel; the score blends a live heuristic with whatever outcomes you log.

1
Product
2
Actor
3
Fit + Learn
1 · Upload your product
A real image — its dominant color is actually extracted and used in scoring.
Category chips set the product category for scoring — still upload a real image above.
Powered by the best

Every top model. One check.

Fit Check’s affinity table is built to generalize across whichever model ends up rendering — Seedance, Kling, Veo, or whatever’s next.

Seedance 2.0 ByteDance · video
Seedance 2 Fast ByteDance · video
Seedance 2 · 4K ByteDance · video
Kling 3.0 Kuaishou · video
Veo 3.1 Google · video
Veo 3.1 Fast Google · video
Runway Aleph Runway · video
Nano Banana Pro Google · image
Nano Banana 2 Google · image
GPT Image 2 OpenAI · image
FLUX.2 Black Forest Labs · image
Recraft 4.1 Recraft · image

Vendor marks are plain lettered placeholders, not the real logos.

Actor library

Two signals, side by side

Today, actors rank by how often they're used. Fit Check adds a second signal: how well they actually convert per category.

Usage rank — placeholder standing in for the volume-based ranking that exists today.
Fit score — live from this browser's logged outcomes. Real data, or an honest “no data yet.”
For actors

Volume-based ranking rewards whoever is already being picked, which is self-reinforcing — the busy stay busy. An outcome-based score is a second ladder: an actor who converts unusually well for skincare can surface for skincare briefs without first winning on raw volume. For licensed actors paid per use, that's a route to being cast on evidence rather than on incumbency.

Speak to every market

70+ languages, one affinity table.

The category × actor-type affinity signal is language-agnostic today — in production it would need to account for regional performance differences too.

Hindi Tamil Telugu Bengali Marathi Kannada Arabic Spanish Portuguese French German Japanese Korean Mandarin Indonesian Vietnamese Turkish Italian + 50 more
Honest accounting

What's real vs what's a stand-in

A prototype is only useful if it's clear which parts would survive contact with production.

Real
Pixel-level color extraction
Your uploaded image is actually read via Canvas getImageData and averaged into an HSL color — genuine computation, not decoration.
Real
Bayesian outcome blending
Logged outcomes shrink the score toward real data using a proper prior-weighted average — a named statistical method, not a naive mean of clicks.
Stand-in
Affinity table & hue-harmony rule
The category/actor affinity numbers are hand-set for this demo. In production these would be learned from real render-outcome data — same architecture, trained weights.
Stand-in
Delivery-tone signal
The third signal in the score, and the thinnest of the three. No audio is recorded, loaded or analysed anywhere in this prototype — it is two typed-in tables: a 0–1 energy value per persona, and the energy each category suits. In production the per-actor value would come from the voice samples the actor library already holds, and the per-category value from how energetic deliveries actually performed. Here it is a placeholder for that, weighted lowest (25%) precisely because it is the least substantiated.
Grounded
Built on HexCoded's real economics
Three things that already ship make this an extension, not an invention:
  • Free failed renders — wasted spend is already treated as a problem worth solving. Just not this kind of waste.
  • Informal per-actor fit tags on the casting site — actor-category fit is already in the product's vocabulary, only unmeasured.
  • API access on Team and Scale — exactly where a pre-flight scoring call would sit.
Creative Studio hand-off

—

—
— —
— your uploaded product
Illustrative preview — not a real render.
Fit score for this pairing
—/10

An actual render would use Veo 3.1 via Creative Studio.

In production this hands off to Creative Studio, which spends the credits and produces the finished video. Fit Check runs first precisely so that spend is informed — what you see above is a layout mock of the pairing, composed in the browser.