How to Photograph
a Problem That Actually Works
The quality of your photo directly determines whether the AI reads your problem correctly. These tips take 30 seconds to follow and make the difference between an accurate solution and a misread.
The 4-Point Pre-Upload Check
Run through this before hitting Upload.
Is it well-lit?
No shadows across the text. Natural daylight or overhead lamp only.
Is it flat & overhead?
Camera directly above the page, not at an angle. Flat page, not curled.
Is the problem in frame?
Full problem visible with a small margin around it. Nothing cut off.
Is handwriting clear?
Printed, not cursive. Each symbol distinct. Dark ink on white paper.
Six Tips for Maximum Recognition Accuracy
Follow these and the AI will read your problem exactly as you wrote it.
Light from above, not the side
Side lighting creates shadows that run across text and make characters ambiguous to the recognition model. The OCR layer sees light variations as changes in ink density — a shadow over a “1” can make it look like a “7”.
Position a desk lamp directly above the page, or photograph near a window with diffuse daylight. Avoid direct sunlight, which creates harsh overexposure that washes out pencil marks.
Common mistake: Holding the phone in one hand and turning on a desk lamp that’s to the side. The hand-held shadow covers part of the problem. Lay the page flat and use the phone’s rear camera with a timer or voice trigger.
Straight overhead, not at a slant
A camera angle of even 15–20° introduces perspective distortion — characters at the far edge of the page appear narrower than those at the near edge. The preprocessing layer corrects mild distortion, but significant angles degrade accuracy, especially for problems with aligned columns (like a system of equations).
Hold the camera directly above the page so the phone is parallel to it. Most phone cameras have a small level indicator that appears in the viewfinder when the phone is horizontal — use it.
Crop to the problem, not the whole page
Uploading a full textbook page when you only need one problem forces the segmentation algorithm to locate the relevant region among multiple candidates. This works, but it introduces unnecessary processing steps and occasional misidentification of which problem you intended.
Crop tightly to the problem — leaving a small margin of white space on all sides — before uploading. On a phone, use the native Photos app to crop. This single step improves recognition speed and reduces ambiguity when multiple problems are on the same page.
Problem 3.3 ← which one?
Problem 3.4 · Problem 3.5
Print clearly — math isn’t prose
Mathematical symbols that look similar — 1 and 7, 0 and O, 2 and Z, × and x — are distinguished by context in prose text but are genuinely ambiguous in equations. Clear printing that keeps each character distinct reduces misidentification significantly.
Specific guidance: close your zeros (circle, not open top); write “1” with a straight vertical stroke rather than a serif flag at the top; use a horizontal bar to distinguish Z from 2; write multiplication as · or parentheses rather than ×, which can be confused with the variable x.
Best result: Rewrite the problem on plain white paper with a dark ballpoint pen before photographing. The contrast and clarity of a rewrite almost always outperforms a photograph of pencil on graph paper.
Open 0 → read as O
Cursive x → read as ×
Screenshots beat photos every time
If your problem comes from a digital source — a PDF textbook, an online assignment, a learning management system — take a screenshot instead of photographing your screen. A screenshot is pixel-perfect: no distortion, no lighting variation, no camera shake, no reflection.
On Windows, use Win+Shift+S (Snipping Tool) to capture just the problem region. On macOS, use Cmd+Shift+4. On mobile, use the screenshot shortcut and then crop. The resulting image will be the highest-quality input you can provide.
| Input Type | Accuracy |
|---|---|
| Digital screenshot | 97–98% |
| Scanned document (300 DPI) | 95–97% |
| Photo of printed textbook | 89–93% |
| Clear handwriting on white | 84–88% |
| Handwriting on graph paper | 68–75% |
| Photo of screen (phone camera) | 60–72% |
One problem per upload for multi-part questions
Multi-part problems — (a), (b), (c) — can be uploaded together, but results are more reliable when each part is submitted individually. The segmentation layer attempts to identify sub-problems, but heavily nested or annotated questions sometimes cause it to group parts incorrectly.
For word problems that span multiple sentences and include a diagram, ensure both the text and the labeled diagram are within the same crop. The solver needs to see the numerical labels on the diagram alongside the question text to set up the correct equation.
Complete Do & Don’t List
Save this for reference when uploading.