Last updated

Sinhala Image to Text

Free Sinhala OCR — the language most vendors skip, measured at 97.5%.

VisionParse converts Sinhala images to editable UTF-8 text free — one of very few OCR services that supports Sinhala at all, built by a Sri Lankan company.

Most Western OCR products do not offer Sinhala; the big cloud APIs bury it or skip it. We measured ours instead of just listing it: 97.48% character accuracy on a printed paragraph including combining vowel signs and a conjunct form (ව්‍යාපාරයට). Sinhala documents are why this company exists, and this page carries the number to prove the support is real.

Last updated: 2026-08-03 · Measured by VisionParse engineering

  • Free · no account
  • JPG · PNG · WEBP · TIFF · BMP · HEIC
  • 3 MB · 3 pages
  • Deleted within 1h
No file yetReady

Drop an image to read it

JPG, PNG, WEBP, TIFF, BMP, HEIC — up to 3 MB. Your file is read on our servers in Frankfurt and deleted within the hour.

5 files a day3 an hour3 MB a file3 pages a filedeleted after 1h

What this tool does, in numbers.

  • ACCURACY

    Sinhala printed text read at 97.48% character accuracy in our measurement, scored against exact ground truth.

  • SPEED

    VisionParse read a Sinhala paragraph in 0.29 seconds on an 8-vCPU server.

  • LANGUAGES

    The Sinhala model reads the full syllabary — base letters, combining vowel signs (පිලි) and joined conjunct forms — into standards-compliant Unicode.

  • LIMIT

    Older orthography and low-resolution newsprint score below our published figure: the loops that distinguish similar letters close up when pixels run out.

  • ACCESS

    The Sinhala image to text tool requires no account, no email address and no payment card.

  • RETENTION

    VisionParse deletes free-tool Sinhala files one hour after conversion.

Where the 97.48% character accuracy comes from.

Every figure on this page traces to a dated run on hardware we name. Nothing here is a vendor estimate.

Measured 2026-08-03

Claim
Sinhala print, including a zero-width-joiner conjunct, read at 97.48% — real support, not a listed language nobody measured.
Method
VisionParse standard recognition tier, single-threaded. Full methodology.
Sample
A 119-character Sinhala document paragraph rendered with Sinhala (Brahmic syllabary)-correct Noto type through Pango (real text shaping), read through the production pipeline with the sin language model, and scored character-by-character against the exact source.
Measured
2026-08-03 on Contabo Cloud VPS 8 — 8 vCPU AMD EPYC, 24 GB RAM, Ubuntu 24.04.4
Limitations
Modern print in a standard typeface; decorative typefaces and old-orthography books will score lower. A single synthetic document per variant is a small sample and is not a substitute for a corpus. These figures describe format behaviour, not general accuracy. A 500-document corpus run is scheduled and will supersede them.
Contact
VisionParse@senithu.lk

What this tool gets wrong

HELD UP

Conjuncts and combining signs read correctly. The test paragraph deliberately included ව්‍යාපාරයට — a form with a zero-width joiner that breaks naive text pipelines — and it came back intact and correctly encoded. Copy-paste into any Unicode-aware editor works.

KNOWN LIMIT

Low-resolution newsprint is the hard case. Sinhala letters are distinguished by small loops and openings; below roughly 30 pixels of letter height those features merge. Photograph documents flat and close rather than at arm’s length.

Three steps, no account.

  1. Drop your Sinhala image

    A scan, a photo or a screenshot — JPG, PNG, WEBP, TIFF, BMP and HEIC are all accepted.

  2. We read it in the right language

    This page's converter runs Sinhala recognition (sin) — not English recognition guessing at Sinhala (Brahmic syllabary) shapes.

  3. Take the text

    UTF-8 text appears with a confidence score. Copy it or download a .txt; the file is deleted within the hour.

What this tool accepts.

PropertyFree tool
Language modelSinhala (sin), Tesseract 5 traineddata
ScriptSinhala (Brahmic syllabary)
Measured accuracy97.48% character accuracy (2026-08-03)
Measured speed0.29 s per paragraph
OutputPlain text, UTF-8
Maximum file size3 MB
Rate limit5 files per hour, per IP address
File retentionDeleted 1 hour after conversion
Processing locationFrankfurt, Germany

When to use the API instead.

This page converts one file now. To convert files from your own software, the same engine runs over HTTP from Starter upward.

# same engine, from your code
curl -X POST https://api.visionparse.app/v1/ocr \
  -H "x-api-key: $VISIONPARSE_KEY" \
  -F "file=@image.jpg"

→ 202 { "jobId": "c0df8e2c…" }

Use the API when

You convert more than a few files, convert on a schedule, or need results inside another system. Batch, webhooks and bounding boxes are API-only.

Frequently asked

Why do so few OCR tools support Sinhala?

Small market, complex script, no commercial pressure — most vendors never trained a model. We are a Sri Lankan company; Sinhala was a founding requirement, not an afterthought, and we publish a measured figure for it.

Does it handle "yansaya" and other conjunct forms?

Yes — our test paragraph included a ZWJ conjunct and it read correctly with proper Unicode encoding, so it renders right everywhere Unicode Sinhala renders.

Can it read mixed Sinhala and English documents?

Yes. Sri Lankan documents mix scripts constantly, and the recognition handles Latin text and digits appearing inside Sinhala pages.

What about old or degraded book scans?

We measured one: a genuinely degraded Sinhala book page read at 94.5% on this free tool, with errors on similar-loop letter pairs. The same scan read at 99.2% through the AI model our paid extraction uses — so for archival material, the paid tier has a measured, not claimed, advantage.

Is the Sinhala image to text tool free?

Yes — 5 files a day with no account, no email and no card. A free account raises the allowance to 100 pages a month.

Do you keep my Sinhala documents?

Free-tool files are deleted one hour after conversion, and no customer file is ever used to train models.

Measured by VisionParse engineering. Method and raw data are published.Updated Benchmark · Method · Data handling