Technology · Computer vision · ML

AI card grading, engineered for accuracy.

SnapGrade is trained on 50,000+ cards with verified PSA outcomes. It's not a chatbot — it's a purpose-built computer vision pipeline calibrated against the same axes professional graders use.

Try SnapGrade free →
Computer vision50K+ training set412 verified returnsOpen track record
14,200+
Cards graded
87%
Match within ±0.5
5,600+
Active collectors
$2.4M
PSA fees saved
How the model works

Four models, one grade.

We don't run one big model — we run four specialists, one per grading axis, and combine their outputs with calibrated uncertainty.

Centering model

Pixel-precision border detection produces left/right and top/bottom centering ratios — far more accurate than the human eye.

Corners model

A convolutional model trained to detect whitening, rounding and corner chips at sub-millimeter resolution.

Edges model

Edge-runner detection of nicks, dings and bevel wear — including holo-edge silvering visible only at high magnification.

Surface model

Detects scratches, print lines, holo scratches and orange-peel surface artifacts using multi-spectral analysis.

Calibrated confidence

Each model emits a confidence score. When combined confidence drops below 70%, you get a refund automatically.

Continuous retraining

Every verified PSA return is added to the training set. The model gets sharper with every batch.

Why a custom AI beats general image models

You can ask GPT-4 or Gemini "what grade is this card?" and get a confident answer — but that answer is unreliable. General models weren't trained on grading and have no calibration against actual PSA outcomes.

SnapGrade is different. We trained four specialized models against 50,000+ cards with verified PSA returns, so each axis (centering, corners, edges, surface) has its own calibrated prediction. The result is a grade we can stand behind — and refund if our confidence is low.

Verified accuracy

Common questions

Questions we hear all the time.

Is SnapGrade just ChatGPT or Gemini under the hood?
No. SnapGrade is a custom computer vision pipeline — four specialist models trained on a proprietary dataset of 50,000+ cards with verified PSA returns. General LLMs are not used for grading.
How does the AI handle edge cases or rare cards?
For very rare prints, our model's confidence drops and we refund the credit automatically. We'd rather not charge than charge for an unreliable grade.
Can I see what the AI is "looking at"?
Yes — every report shows per-aspect confidence and visible markers on the card image where the model identified centering, corner wear, edge issues or surface defects.
Is my card data used to train the model?
Only with opt-in consent. By default, your card photos are kept private to your account.
How often does the model update?
Continuous training — every new verified PSA return becomes part of the dataset. Major model updates roll out monthly.
Ready when you are

Run a card through the model
right now.

Sign up free, get 2 free credits, and see what a calibrated AI prediction actually looks like — with subgrade-level confidence, not a chatbot guess.

Get 2 free grades → Contact us
On signup
+2 credits
No credit card · No expiry