What’s new in FindFlower.

New features

• Identify flowers from a link, including automatic YouTube thumbnail extraction.
• Developer preview: direct-endpoint documentation and gated API keys via secure sign-in.

Model improvements

• New Vision Transformer model, now served from a secure hosted endpoint instead of downloading to the browser. Your photo is sent for processing and only the result comes back — this changes the privacy model, so see the privacy page.
• Expanded coverage to 116 botanical classes.
• Steadier confidence scoring, with clearer separation between close look-alikes.

Interface

• Refined result panel with animated confidence and top alternatives.
• Live Wikipedia context, with reference image and full-article link.

Fixes

• Resolved model-loading issues on browsers with strict cache settings.
• Corrected preview handling for large photos.

New features

• Live camera capture, alongside image upload.
• Feedback prompt after each result to guide future improvements.

Model improvements

• Fully in-browser inference — images no longer leave the device.

Fixes

• Improved camera permission handling across browsers.

The first public Beta

• Core identification from an uploaded photo.
• The calm, single-workspace interface FindFlower is built around.
• Initial model covering common garden flowers.