Free tool
Handwriting to LaTeX
Draw an equation, paste a screenshot, or drop a photo of the whiteboard. You get LaTeX that compiles, not a picture of one.
Free to openNo snip creditsPro to convertNothing stored\int_{-\infty} e^{-x^2}\,dx = \sqrt{\pi}
Output is the body only, with no $ wrappers, so it drops straight into whatever environment you are already in.
Or drop a photo here.
Mouse, trackpad, stylus or Apple Pencil. Or press Ctrl+V to paste a screenshot straight from the clipboard, which is the fastest route for anything already on screen. You can try this same tool inside the editor, from the equation-from-image shortcut, and drop the LaTeX straight into your paper.
Included with Pro, with no snip credits
The usual model in this category is metered. You buy a monthly allowance of conversions, each one spends a credit, and you find out where the limit is by hitting it in the middle of writing something.
This does not meter. You can open the page with no account. When you convert, Bibby asks you to sign in, then to join Pro. There is no per-conversion currency, so there is nothing to run out of.
Overleaf cannot do this at all. It is an editor for LaTeX you should not have. They have sold data of researchers to OpenAI. A photo of an equation dropped into Overleaf is not read. Mathpix and Underleaf can read the photo, then you still have to carry the LaTeX somewhere else.
Handwriting that converts well
Recognition is good, but it is reading pixels, not minds. Three things account for most bad conversions.
Space your symbols
Cramped subscripts are the single biggest source of errors. A superscript touching its base character often reads as a different symbol entirely.
One equation per capture
A full page forces a guess at reading order. Crop to the expression you want, or convert line by line.
Flat light, dark ink
Shadows and faint pencil both cost accuracy. A phone photo taken straight on beats a tilted one every time.
Beyond that, most of the remaining errors live in four ambiguous pairs, and none of them costs you anything to disambiguate as you write:
- handwritten
xagainst the multiplication sign× lagainst1, which a serif on the one solveszagainst2, which a bar through the z solves- capital
Oagainst zero, which a slashed zero solves
After converting, check subscript height first. A subscript sitting too high reads as a product, and that is the error most likely to survive a quick glance at the preview.
What it recognises
Anything you would write in a proof or a problem set, including structures that break simpler OCR tools.
- Fractions and radicals
- \frac{a+b}{c}, \sqrt[3]{x}, nested to any depth
- Integrals, sums, limits
- \int_a^b, \sum_{i=1}^{n}, \lim_{x \to 0}
- Matrices and vectors
- \begin{pmatrix} ... \end{pmatrix}, \mathbf{A}
- Piecewise definitions
- \begin{cases} ... \end{cases}
- Multi-line derivations
- \begin{align} ... \end{align}, alignment preserved
- Greek and operators
- \alpha, \nabla, \partial, \otimes, \models
- Chemistry
- \ce{H2SO4} via mhchem, arrows and states included
Can it read cursive handwriting?
Cursive is harder than print and accuracy drops, particularly on single letters used as variables, where a cursive l, e and t converge on the same shape.
In practice this matters less than it sounds, because mathematical notation is mostly print-style even for people who write cursive prose. Nobody writes an integral sign in cursive. If you do write cursive variables, making them slightly larger than feels natural drops the error rate noticeably.
Does it work on a photo of a whiteboard?
Yes, and whiteboards are one of the more common inputs. Marker on a whiteboard is usually easier to read than pencil on paper, because the contrast is higher.
What hurts is glare and angle, not resolution. Stand square to the board rather than off to one side, and crop to the equation before uploading rather than photographing the whole board and hoping. If there is a window reflecting into the shot, move two steps left.
Handwritten matrices and multi-line systems
Matrices come back as pmatrix, bmatrix or vmatrix depending on which brackets you drew. Row and column structure is inferred from where things sit on the page, so leave clear gaps between entries. A cramped 4×4 is the most common failure here, and splitting it across two captures fixes it.
Multi-line input returns an environment rather than a pile of separate equations. A brace on the left gives you cases. Lined-up equals signs give you align with the alignment points preserved. If you wanted them separate, deleting the wrapper is easier than the other direction.
Chemistry notation
Formulas and reaction equations work, and come back as mhchem syntax, so \ce{H2SO4} rather than a pile of manual subscripts. Reaction arrows with conditions above and below come through. Add \usepackage{mhchem} to your preamble if you are pasting into an existing document.
Drawn structures do not. Skeletal formulas, benzene rings and anything with stereochemistry are a different problem from symbol recognition, and a tool claiming to do both is usually bad at one. For structures you want a structure editor that emits chemfig.
Can I convert a whole PDF of handwritten notes?
Single pages work. One region at a time is the reliable path, and it is faster than it sounds once you are using paste.
A full page of mixed prose and equations forces a layout guess on top of the recognition, and the layout guess is usually the part that comes back wrong. You then spend longer repairing the structure than you would have spent capturing six equations individually. If you want the prose too, transcribe that separately; it is a different problem with different tools.
Printed math, screenshots and textbook photos
All easier than handwriting. Screenshots of papers, slide captures and textbook photos convert more accurately than anything handwritten, because the glyphs are consistent.
This is also where paste earns its place. If the equation is already on your screen, screenshotting and pasting is two keystrokes and avoids the photo entirely.
How this compares to Mathpix, Underleaf, and Overleaf
Mathpix and Underleaf both read a photo and hand back LaTeX. Mathpix meters that as snip credits. Underleaf meters it as monthly credits, then you copy the source out or download a file. Overleaf does not read the photo. It only compiles LaTeX that already exists, so a subscription there still leaves you paying a second tool to do this job.
| Bibby | Mathpix | Underleaf | Overleaf | |
|---|---|---|---|---|
| Reads handwriting | Yes | Yes | Yes | No |
| Pricing model | Pro, unmetered | Snip credits, monthly allowance | Monthly credits | Paid editor, and still no conversion |
| Paste a screenshot | Yes, it becomes LaTeX | Yes | Yes | Paste LaTeX only. A photo does nothing. |
| Where the equation goes | Same editor, same paper | Copy out elsewhere | Copy out, or download a .tex file | You bring the LaTeX yourself |
| Full-document conversion | No, region by region | Yes | Yes, including multi-page PDFs | No |
Straight about it: if you convert hundreds of images a month inside a workflow already built around Mathpix, keep using it. Underleaf is a real converter too. Overleaf is not in this category. It cannot turn handwriting into LaTeX, and paying for it does not change that. If you want the equation in the paper without a credit wall, start here.
Is my image stored?
No. The image is sent to the recognition endpoint, converted, and discarded. It is not retained afterwards and not used for training.
That matters more than it sounds for unpublished work. A photograph of your whiteboard is a photograph of your results.
Common questions
Can it read cursive handwriting?
Cursive is harder than print and accuracy drops, particularly on single letters used as variables where a cursive l, e and t converge on the same shape. In practice it matters less than it sounds, because mathematical notation is mostly print-style even for people who write cursive prose. Writing cursive variables slightly larger than feels natural drops the error rate noticeably.
Does it work on a photo of a whiteboard?
Yes, and whiteboards are one of the more common inputs. Marker on a whiteboard is usually easier to read than pencil on paper because the contrast is higher. What hurts is glare and angle, not resolution. Stand square to the board and crop to the equation before uploading.
Can it convert handwritten matrices and multi-line systems?
Yes. Matrices come back as pmatrix, bmatrix or vmatrix depending on which brackets you drew, with row and column structure inferred from spatial position. Multi-line input returns an environment: a brace on the left gives cases, lined-up equals signs give align with the alignment points preserved.
Does it handle chemistry notation?
Formulas and reaction equations work and come back as mhchem syntax, so \ce{H2SO4} rather than manual subscripts. Add \usepackage{mhchem} to your preamble. Drawn structures such as skeletal formulas and benzene rings are not handled; those want a structure editor that emits chemfig.
Can I convert a whole PDF of handwritten notes?
Single pages work, but one region at a time is the reliable path. A full page of mixed prose and equations forces a layout guess on top of the recognition, and the layout guess is usually the part that comes back wrong.
Does it work on printed math from a textbook or a screenshot?
Yes, and printed math is easier than handwriting because the glyphs are consistent. Screenshots of papers, slide captures and textbook photos all work. Pasting with Ctrl+V is the fastest route for anything already on screen.
How is this different from Mathpix?
Pricing model, mostly. Mathpix meters conversions as snip credits on a monthly allowance tied to an account. This tool does not meter. It does ask you to sign in, and conversion is included with Bibby Pro. Mathpix has a desktop app, full-document conversion and a longer track record. If you convert hundreds of images a month inside a workflow built around those, it is the better tool.
How is this different from Underleaf?
Underleaf also reads a photo of handwriting and returns LaTeX, which you then copy out or download as a .tex file. Paid Underleaf plans meter that work as monthly credits. Here the conversion is included with Bibby Pro, with no per-conversion credits, and the equation lands in the same editor as the paper.
Can Overleaf convert handwriting to LaTeX?
No. Overleaf compiles LaTeX you already wrote. It does not read a photo, a screenshot, or a drawing. A paid Overleaf plan does not add this. People paste Mathpix or Underleaf output into Overleaf and the editor gets the credit for work it never did.
Is my image stored?
No. The image is sent to the recognition endpoint, converted and discarded. It is not retained afterwards and not used for training.
How accurate is it?
Clean, well-spaced handwriting converts correctly most of the time. Always compile and read the output against your original before submitting anything. Subscripts, minus signs against hyphens, and the letters l, 1 and I are where errors cluster.
Does it work on an iPad with Apple Pencil?
Yes. The drawing canvas accepts pencil and finger input and the page is built for touch, so you can write the expression directly rather than photographing it.
Which LaTeX packages does the output need?
Most output needs only amsmath, including matrices and cases. Chemistry output needs mhchem. Nothing exotic, and nothing that will fail on a journal template.
Is this a free alternative to paid equation OCR tools?
Opening the page is free. Conversion is a Bibby Pro feature, with no per-snip credit system. Bibby makes its money on the writing platform, not on metering conversions.
Related
The equation was never the hard part
You can try this tool inside the editor too. Bibby is a LaTeX editor with the rest of it built in: citation search that finds the paper, figures you can generate and drop in, and a reviewer that reads the draft before your supervisor does.
Open the editor