Accessibility here is about getting information out of a picture and into a format that can be read, enlarged or captioned. The chain is OCR the photo or the PDF, upscale a screenshot that is too small, add captions to a video, and keep a clean copy of the text. Nothing here speaks, and that is the boundary of the page.
Order matters. Read the text first, because OCR is the step that turns an inaccessible image into something a screen reader can handle. Upscale after OCR, not before, since a bigger image does not improve recognition much and costs time. For video, decide between a caption file and burned-in captions: the file keeps the original quality and lets the viewer control size and colour, while burning makes the captions impossible to switch off or restyle. Keep the extracted text as a plain text file, because that is the copy that works with every assistive tool.
1. Read the text out of a photo
Image to Text runs Tesseract compiled to WebAssembly on your own machine, so the photo is never uploaded. The first run downloads about 3.7 MB of engine plus a 2.8 MB English language file, then caches both. English is built in and eleven more languages can be selected, including Chinese, Japanese, Korean, German, French, Spanish, Russian and Arabic. You can drop several images at once and they are read one after another, with a filename line between results.
Accuracy is high on clean printed text, a straight-on photo or a screenshot, and poor on handwriting, stylised logos and heavy shadows. That is not a small caveat: a menu photographed at an angle or a page with a strong crease will produce mistakes, and every one of them has to be proofread. The output is an editable text box, so copy it into the app you actually read with. There is no voice output here.
Imagem para texto
OCR a photo, screenshot or scan locally, in twelve languages, with no upload.
2. Read a scanned PDF as text
PDF to Text handles both kinds of document. A PDF with a real text layer is read exactly and instantly; a scanned page is rendered and recognised with OCR instead. The result panel reports how many pages came from each route, which tells you where mistakes are likely. English is built in and eleven more languages can be selected, and everything runs locally, so a medical letter or a bank statement is not uploaded just to be read.
To hear the result, copy the text into a screen reader or a text-to-speech app, because there is no speech synthesis on this site. Two limits are worth planning around: handwriting is unreliable, and an encrypted PDF has to be unlocked before it can be read. The output is plain text without layout, so a table becomes lines rather than cells and a two-column page interleaves the columns.
PDF para texto
Read the text layer or OCR the scans, locally, in twelve languages.
3. Enlarge a screenshot that is too small
AI Image Upscaler runs an ESRGAN model locally and offers 2x, 4x, 6x or 8x, with 2x and 4x in one pass and 6x and 8x chaining two passes, up to 20 images per batch. It cleans compression artifacts and rebuilds edge contrast, which can make small interface text easier to read on a large screen. Output is capped at 8192 pixels on the long edge and 24 megapixels in total, and nothing leaves the machine.
This is enlargement, not a font-size control. If a screenshot is small but sharp, upscaling gives you a bigger, still crisp image; if the text is blurred by compression, it may only look smoother. A letter that was never resolved cannot be recovered, and the tool does not reflow a document or change its type size. Export PNG for on-screen reading, and keep the original for reference.
Ampliador de imagens com IA
Local 2x to 8x for a screenshot that is hard to read.
4. Add captions to a video
Add Subtitles burns an SRT, VTT, ASS or SSA file into the video, with controls for size, colour, outline and position. Subtitle Converter rewrites a subtitle file into a format a player can load, and supports a global time offset. If you already have captions, the sidecar file is the better choice: the player can scale and recolour it, and the video is not re-encoded.
The burner has one built-in Latin font, so non-Latin scripts and custom styling are not covered, and it re-encodes the whole clip. Captions also carry dialogue, not sound effects: a description of a doorbell, a phone ringing or a knock has to be written into the subtitle file by hand. That is the difference between subtitles and captions for a deaf viewer, and no tool here writes audio descriptions.
Adicionar legendas
Burn a caption file into the video, or convert it for a player instead.
5. Draft captions from the audio
Video to Subtitles turns a recording into a transcript with timecodes and speaker labels, and you can correct names and wording before downloading SRT, VTT or plain text. That draft is the practical starting point when no caption file exists, because writing captions from scratch in an unknown language is slow and error-prone.
This is the one step that uploads. The file goes to a server and is forwarded to Volcengine AI MediaKit, caps at 30 MB, recognises Simplified Chinese and English only, and a free account receives 5 credits a day. For a private conversation or a medical recording, that upload is a privacy decision, not just a convenience. De-identify the recording first, or caption it by hand.
Vídeo para legendas
Draft a caption file from the audio. Uploads the file and uses credits.
What this chain cannot do
- It does not speak. There is no text-to-speech and no screen reader; the output is text you copy elsewhere.
- No image descriptions. Nothing here describes a photo for a blind reader; it only reads text that is already printed in it.
- Handwriting is unreliable. Printed OCR covers twelve languages; a handwritten note does not.
- Upscaling is not a font-size control. It enlarges pixels and cannot recover a letter that was never resolved.
- Captions are Latin-only when burned in. Non-Latin scripts need a sidecar subtitle file.
- No audio description. Sound effects and scene changes are not captioned automatically.
- Transcription uploads. 30 MB, Chinese and English only, and a privacy decision for personal recordings.
- Encrypted PDFs block reading. Unlock the file first, and keep the text as plain text for other tools.
- No contrast or reading-mode changes. The page cannot restyle a document for dyslexia or low vision.