Hello, blindGram Assistant version 0.1 is released.

The Assistant is a separate, optional app. It adds three things to blindGram: reading the text inside an image, describing a photo, and turning a voice message into text. One install serves both blindGram and blindGram Plus. Everything except the AI description works offline, with no account and no Premium. It works only while "blindGram Accessibility" is on.

Install, update, remove

Open blindGram Settings, then "blindGram Assistant" (the last entry, after Menu), and choose install; Android asks for confirmation and offers the unknown-sources setting if it is still off. An interrupted download continues from where it stopped. The installed version, "Check for updates" and remove are at the end of the same screen.

Where the features are used

All three are reachable in four places: the message menu on the message itself, a selected message under "blindGram Assistant", "More options" in the photo viewer, and the Assistant button while choosing an image to send. Every result opens in a dialog the screen reader reads at once, and can be copied.

Image to text

Reads the text inside a photo, screenshot or scanned page: "Extract text from image". Persian and English are included; eighteen more can be downloaded in its settings. "Recognition languages" chooses which of the installed languages are used, and more than one can be ticked — Persian alone reads Persian most accurately, while adding English also reads Latin text but may misread a Persian word. The larger the text is in the picture, the better it is read.

Describe image

Says what is in a photo that has no text to read: "Describe image". The answer comes in two parts. First, object names from a model on the device: immediate, offline, nothing to set up, in Persian. Then, if a description service is set up, full sentences replace that answer in the same dialog and are read again; the dialog says while it is waiting, and says why if the service cannot answer.

That service is optional and off until switched on, because it sends the photo away; switching it on shows a one-time notice naming where. In its settings: Service (ready-made choices, and OpenRouter, which works on a free account, is the default, or Custom for any address), Key from your own account, Model from a list, and at the end Description language, which follows the app's language until another is chosen. A description is remembered, so hearing the same photo again costs nothing and does not send it a second time.

Voice to text

Writes out a voice message or a round video, offline and without Premium: "Convert voice to text". Speech models are large, so none is included — the first attempt offers the download for the chosen language with its size, then carries on with the same message. Twenty-three languages are available, Persian and English first. Only one is used at a time, so the language row appears once a second model is installed. Expect the sense of what was said rather than every word exactly.

Notes

Every download shows a progress bar and speaks the percentage. The Assistant needs no storage permission. In a description language other than Persian, the first on-device part still names objects in English.

@blindGram_Assistant