Android Available Audience: Blind

Owli-AI Assist

Describe scenes, recognize text and ask follow-up questions; read or listen to answers.

Notice: this page was machine-translated and is pending editorial review.

Good to know

AI descriptions and recognized text can be incorrect or incomplete. Follow-up answers refer to the saved image, not to changes since it was taken. Check important information; do not base safety-critical decisions solely on the app.

German promotional graphic for Owli-AI Assist with owl artwork and references to AI image description, OCR, and follow-up questions

Core features

Describe scenes and recognize text

Profiles offer a short overview, detailed description or text recognition (OCR) for documents, signs and displays.
Choose processing mode

The Owli backend is the default. Advanced settings offer your own OpenRouter key and local Gemma 4 on suitable devices.
Follow-up questions with image context

Ask questions about a captured scene. Direct OpenRouter mode supports additional image attachments.
Auto-scan

Supported profiles analyze new camera images at regular intervals.
Speak questions and hear answers

Dictate questions and listen while answers arrive. Repeat and stop controls are available.
TalkBack operation

Large controls and focus management support screen reader use.
Report a problematic answer

Report answer text, category and an optional comment; images are excluded.
Photo to song video

The experimental feature creates a short English song from an image, shareable as a video.

Privacy

Operating mode: Cloud

The camera preview stays local. By default, Assist sends AI requests through the Owli backend after your action; your own OpenRouter key and local Gemma 4 are advanced options.

Backend follow-ups resend the original image. The song feature may use external AI/music services. Reports without images are currently retained for 30 days.

Go to privacy page

Install directly on Android

Install the app from Google Play or scan the QR code if you are viewing this page on a desktop computer.

Install on Google Play View all Owli-AI Android apps

QR code for the Google Play page of Owli-AI Assist

Open it quickly on your phone

Scan the QR code with your phone to open the store page directly.

System requirements

Android 10 or newer
Camera and microphone
Internet connection for backend and OpenRouter
Local Gemma 4 requires a suitable device, sufficient RAM and about 2.6 GB of model storage.

Details and usage

Capture an image and ask follow-up questions

Assist is designed for blind and severely visually impaired people. It provides additional information about scenes, text and objects. It does not replace personal assistance or your own checks in safety-critical situations.

Point the camera at the scene and select New scene.
The selected processing mode analyzes the image. Read the answer or listen to speech output.
Ask questions such as ‘What does the sign say?’ or ‘Read the lower section’. Backend mode sends the original image again for each follow-up.
Reset the active scene before taking a new image. Additional images are available in direct OpenRouter mode.

Choose description or text recognition

Profiles define the task: a short overview, a detailed description or faithful reading of documents, signs and displays using text recognition (OCR). The default provides a short description suitable for speech output.

Some profiles support auto-scan, analyzing new camera images at regular intervals.

Speech and TalkBack

A short tap on the microphone can send the recognized question directly; a long press allows dictation followed by editing. Speech output can begin while the answer is arriving. You can repeat the last answer or stop playback.

Large controls and focus management support TalkBack. Profile switching is restricted during a request or while an active scene is locked. Portrait and landscape layouts have adapted controls; indicators show ongoing capture, speech recognition and requests.

Turn a photo into a song

The experimental feature creates a short English song from a captured or imported image. The image and prompt are sent through OpenRouter to Google (Lyria 3 Clip Preview) for this purpose. You can share a video of the image and audio with other apps through Android sharing. The Owli-Song page provides examples and three usage steps. Results can vary.

Advanced settings for experienced users

AI requests use the Owli backend by default, without your own API key. Advanced settings offer alternatives:

Your own OpenRouter key (BYOK): save a key or import it via QR code and send requests directly to OpenRouter. Key information can show usage, limits and expiry.
Local Gemma 4: download and select a model on a suitable device. The app checks device capability, storage and download status. Local inference requires a ready model.

Profiles also control image resolution, compression, model, maximum answer length and incremental output (streaming). These technical settings are not required to get started.

Report a problematic answer

Report AI answer sends the answer, a category and an optional comment to the Owli backend, together with technical metadata. Images and screenshots are excluded. Reports are currently retained for 30 days for review.

Privacy and processing

The camera preview stays local. An active AI function triggers transfer of image and question data. Backend follow-ups resend the original scene image. Direct OpenRouter mode sends the image, question and any additional image attachments directly to OpenRouter.

The song feature may use the Owli backend and external AI/music services. Even with local Gemma inference, other functions such as help pages and feedback can access the internet. Settings, profile selection and caches are stored locally; your own OpenRouter key is encrypted locally. See the Assist privacy policy for further information.