Seer – Open-source local AI image descriptions for screen readers (no API key) (github.com)

Seer, a new open-source browser extension developed by Recursia Lab, has been launched to provide local AI-generated image descriptions for users of screen readers, eliminating the need for an API key or cloud connectivity. This tool uses the PaliGemma2 AI model to automatically describe images that lack alt text, catering specifically to individuals with visual impairments who navigate the web using screen readers like NVDA, JAWS, or VoiceOver. By processing images directly on users’ computers, Seer ensures that privacy is maintained, as no data leaves the local environment, facilitating a more accessible online experience. The significance of Seer lies in its potential transformation of web accessibility. Many websites fail to provide descriptive alt text, leaving visually impaired users to encounter inaccessible content. With Seer, users can effortlessly interpret visual information, even when internet access is limited or unavailable. Technically, the extension is lightweight, requiring about 3 GB of RAM and utilizing a 3B vision-language model for various tasks, including image description and OCR. This empowers individuals in low-resource settings, reinforcing the notion that accessibility tools should be free and universally available. The tool is licensed under Apache 2.0, encouraging further collaboration and development in the AI accessibility space.