How to extract text from screenshots

A screenshot can preserve information while making it difficult to reuse. Optical character recognition, or OCR, turns the visible characters into editable text. TinyModel runs PP-OCRv6 Small in the browser, including text detection and recognition.

Prepare an upright, clear image

Choose a JPG, PNG, or WebP. Rotate sideways images with the workspace control before processing. Text should be large enough to read, with as little blur and glare as possible. Crop away unrelated areas when you only need one section.

The first version accepts images, not PDF files. Export a PDF page as an image first if you need to recognize a scan. Native PDF text extraction and searchable PDF creation are separate workflows planned for later.

Extract and review

1. Open Image to Text.
2. Choose the image and check the preview.
3. Select Extract text. On the first run, the browser downloads the detection and recognition models.
4. Read the output next to the original and correct any errors.
5. Copy the text or download a TXT file.

PP-OCRv6 Small uses a unified multilingual recognition model. You do not need to choose a language pack. Supported character coverage includes Chinese, English, Japanese, and multiple Latin-script languages. This does not mean every language or writing system is supported equally.

What happens locally

The detector locates likely text regions. The pipeline maps those locations back to the source image, crops each region, and passes it through the recognizer. A character dictionary converts the model output into text. The lines are then assembled in a basic reading order.

The two model weight files total around 31 MB; the browser runtime and supporting assets are additional downloads. The image and recognized text are not sent to an OCR server.

Check the details that matter

Compare names, dates, quantities, decimal separators, and punctuation carefully. Common confusions include the letter O and digit 0, the letter l and digit 1, and punctuation at low resolution. Mixed scripts can also need correction.

A blank result does not prove that the image contains no text. The text may be too small, rotated, low-contrast, or outside the recognizer's character coverage. Try a clear crop of the relevant region.

Layout has limits

A plain-text export does not preserve page design. Multiple columns can be placed in an unexpected order. Tables can become a list of lines rather than rows and columns. If you need a spreadsheet, review and restructure the text yourself; this version does not promise image-to-table conversion.

Handwriting, curved labels, and heavily distorted documents are more difficult than ordinary upright printed text. TinyModel provides editable output so that review remains part of the workflow.

Keep sensitive information local

The processing workspace does not include analytics or advertising scripts. Text remains in the current page until you copy, download, replace the input, or close the workspace. It is not saved to an account or a server-side history.