PIXELOKAY MATH LAB · STANDARD / NO AI BY DEFAULT

AI Math Equation Recognizer

Upload or draw a formula, select its boundaries, and turn it into LaTeX with optional local AI. Correct every symbol, preview typeset notation, and export the math you need.

✎ Editable LaTeX∑ Accurate MathML rendering↓ SVG & PNG export
EQUATION WORKSPACE

Math Equation Recognizer

1. Import or draw   2. Select expression   3. Recognize   4. Edit & export

Image & drawing canvas

Start with a file, freehand drawing, or the sample.

Your equation goes here.Drop a photo or start writing with Draw ink.
Blank drawing canvasNo expression selected

Equations & editable LaTeX

Select an equation to review it and export.

0 equations

Equations will appear here. You can also add a blank LaTeX equation without an image.

Math templates & symbolsNo AI required · selecting text wraps it

LaTeX structure helper

Quick symbols
Rendered equationAdd an equation to preview it.
—
MathML source
Graphic export settings
BUILT FOR STUDENTS, TEACHERS & RESEARCHERS

Readable equations, editable source.

Is recognition really on-device?

Yes. The optional free model and runtime are downloaded to your browser. The formula image is used for local inference, not submitted to an API. Models are large and may not work on every device.

Can I work without AI?

Yes. Drawing, crop selection, LaTeX editing, MathML conversion, SVG/PNG exports and project save/reopen work without recognition downloads.

How accurate is handwritten math?

It varies with symbols, handwriting clarity and formula length. Always correct the text before using it in assignments or papers.

Does it handle entire math pages?

Each image can contain several formulas, but select each expression separately. This is not a full textbook or PDF layout recognizer.

Using AI Math Equation Recognizer

Import an image, draw ink or enter LaTeX manually. Select an equation region, optionally recognize it and correct every symbol before preview/export. Save the project to preserve images, equations and editable strokes.

Recognition downloads model/runtime assets and needs review; it is not equation solving or proof of mathematical correctness. The specialized handwritten model remains disabled pending licensing review. No PDF/full-page layout recognition is supplied. MathJax renders supported notation locally; image/equation/stroke/project budgets and undo-scope limits apply. Projects embed potentially sensitive source images.

References: TexTeller model · MathJax

Recognize a selected equation, then correct its LaTeX

Import an image, draw ink or enter LaTeX manually, select one equation region and optionally request TexTeller recognition. Correct the recognized symbols, inspect the bundled MathJax preview, and export supported TeX, MathML, SVG, PNG or project files.

Example workflow

  1. Crop or select one printed expression, or draw it in the workspace, and check that the selection includes the full equation.
  2. Request optional recognition where appropriate, then compare every symbol, exponent and delimiter with the source before accepting the LaTeX.
  3. Review the MathJax rendering and export the needed representation; save the editable project if you need its image, equations or strokes.

Authored example: Use an expression containing a fraction and a superscript. Verify the numerator, denominator and exponent one by one; a clean preview only shows that MathJax rendered the entered notation.

Limits and model handling: TexTeller runtime/model assets download for recognition; MathJax preview is bundled locally. Recognition is not equation solving or proof of mathematical correctness. Select expressions individually rather than expecting full-page textbook layout recognition. The specialized handwritten model remains disabled pending licensing review. Projects embed source images and are subject to image, equation, stroke and project budgets.

AI Math Equation Recognizer questions

Does recognition prove that an equation is mathematically correct?

No. It proposes notation from an image region. Correct the symbols and independently verify any mathematical result.

Can it recognize an entire scanned textbook page at once?

No. Select individual equation regions. It is not a full-page PDF or textbook layout recognizer, and the specialized handwritten model remains disabled.

Related tools

LOCAL AI SETUP

Set up AI on this device

First, a model needs to be loaded on your device

AI Math Equation Recognizer uses the specialized TexTeller ONNX image-to-LaTeX model via Transformers.js. Model/runtime files download only after you choose Download & Load. The cropped formula stays on this device during inference; model files come from external model/CDN hosts.

  1. 1. Check device
  2. 2. Download / cache
  3. 3. Prepare formula model
  4. 4. Recognize formulas
TexTeller ONNX (onnx-community/TexTeller-ONNX) · Transformers.js 3.8.1 · WASM/CPU · image-to-LaTeX. Recognition produces editable suggestions; it does not solve equations or prove mathematical correctness.

Device & model resources

Measurements may be approximate
Device RAM / model memoryChecking…

Live RAM and model working-memory use unavailable.

Checking browser WASM and Worker support…

Browser website storageChecking…

Site-wide estimate includes all same-origin PixelOkay tools, and is not RAM.

ONE ACTION · ALL RECOGNIZED AI MODEL DOWNLOADS

Free AI model storage across PixelOkay

Scan this website origin for WebLLM and Transformers.js model caches and recognizable model files, including other PixelOkay tools. Only positively identified model artifacts in Cache Storage can be removed here, including those cached by other tools. Inaccessible IndexedDB, OPFS, projects, drafts and other user data remain untouched.

Run a scan to identify safely removable model artifacts. Unknown browser-model stores cannot be deleted safely.

Checking model cache does not download files. A complete cache is confirmed only by successfully loading the model; partial entries are not treated as ready.

AI recognition is optional. Drawing, manual LaTeX entry, MathJax preview and exports work without any model download.

The first formula inference may be slower due to model initialization. Live free RAM is not exposed by ordinary browser APIs.

Having trouble downloading or loading AI?

First try Refresh Storage & Cache and Retry Setup. Check internet connectivity, runtime/CDN access and device memory. A failed worker import or unsupported WASM is not necessarily a storage problem.

Do not clear all website data: that can remove saved work across PixelOkay. Only safely identified model assets are considered here.

Technical details and diagnostics

Transformers.js @3.8.1 loads onnx-community/TexTeller-ONNX in WASM mode. Identical model assets in the same origin and cache provider can be reused; partial cache entries are not verified complete. Unknown caches, browser databases and saved projects are excluded.

CROSS-TOOL MODEL CACHE CLEANUP

Delete recognized model downloads?

Other PixelOkay AI tools may need to re-download model files. Close active AI tabs first. Only positively recognized Cache Storage model artifacts are targeted. Unknown caches and IndexedDB are untouched; never clear general website data without exporting work.