Two photos in
You upload the front and the back of the card. The back matters: PSA grades it, and edge and corner wear often show there first.
A plain explanation of what goes in, what comes out, how confidence works, what a photo cannot show, and what we do and do not claim about accuracy.
Updated
Everything below describes what the tool does today. If something is not listed, we are not claiming it.
You upload the front and the back of the card. The back matters: PSA grades it, and edge and corner wear often show there first.
Centering, corners, edges and surface are each scored on PSA's 1–10 scale. Each axis is judged on the most severe defect visible on it, not on an average.
Every axis comes back high, mid or low confidence. If glare, blur or a sleeve hides an area, the axis is marked low instead of guessed.
The overall prediction uses only grades PSA issues: whole numbers from 1 to 10, never a 9.5. The report names the axis that limits the grade.
The grader reads the card's era from its design and copyright line. Vintage cards rarely reach a 9 or 10, so predictions reflect aging.
It predicts condition from photos. It cannot tell you whether a card is genuine.
| Limit | Why it matters | What to do |
|---|---|---|
| Faint surface flaws | Hairline scratches and shallow dents often need raking light to see | Tilt the card under a lamp before you submit anything predicted 9 or 10 |
| Glare on foils and chrome | Reflections hide detail and lower confidence | Re-shoot with soft light; see how to photograph cards for grading |
| Authenticity | A condition prediction says nothing about whether a card is real | Authentication is a grading company's job |
| Borderless and full-art cards | No clear border to measure centering from | Read the confidence on centering and measure by hand if it is low |
| Photo quality | Blur, tilt and low light change the result | Photograph straight on, sharp, in even light |
SnapGradeAI does not publish an accuracy percentage. We removed earlier figures from the site because they were not backed by real PSA results, and we will not quote one until it is. Users can report the real PSA grade (cert number) of a card they submitted, and those reports are the data we are collecting. Until then the honest signal is the confidence level on each axis, and your own test: grade a card whose PSA grade you already know.
In November 2026 the grading was recalibrated toward PSA's real distribution. In an internal check on 41 test cards, the share predicted 9 or higher fell from 95% to 27%. That was a small internal check, not a validation study.
More on what the AI can and cannot see: Pokémon card grading AI. Questions or a result that looks wrong? Contact us with the card photos and what PSA graded it.
Sign up free, 1 free credit, and compare the prediction with the real grade.