How to Search Text Inside Images and Videos on Mac
TL;DR
macOS has basic OCR for images, but it fails with complex searches and ignores video completely. Learn how local AI tools use Florence-2 and Whisper models to make your entire media library semantically searchable.
Our hard drives are no longer just filled with Word documents and PDFs. Today, a massive portion of our digital lives consists of media: screenshots of receipts, photos of whiteboards, recorded Zoom meetings, and downloaded video tutorials.
But there is a major problem: traditional file search cannot "see" or "hear."
If you have a screenshot named Screenshot 2026-05-12 at 10.42 AM.png, your Mac's Spotlight search has no idea what is inside that image unless you manually rename it. The same goes for hours of unlabelled video recordings.
Here is a deep dive into how you can finally search the actual text and contents inside images and videos on your Mac using local AI.
The Limitations of Built-in macOS Features
Apple has introduced some excellent features in recent years, like Live Text, which allows you to highlight and copy text from images in Preview or Photos. Spotlight can occasionally search this text.
However, it is limited: - Exact Matches Only: It still relies on keyword matching. If the text is slightly blurry or you search for a synonym, it fails. - Visual Concepts Ignored: Live Text only reads text. If you want to search for "a photo of a golden retriever on the beach," Live Text cannot help you. - Video is Invisible: Spotlight does not transcribe local video or audio files. The spoken words inside your media are completely unsearchable.
The AI Solution: Multimodal Semantic Search
To truly make your media searchable, you need an application that employs multimodal AI models. These are AI systems capable of processing different types of data (text, images, audio) and mapping them into a shared semantic space.
This is the core technology powering Dhito.
Searching Images with Dhito (Powered by Florence-2)
Dhito uses a highly optimized, local vision model (based on Microsoft's Florence-2 architecture) running on your Apple Silicon.
When you point Dhito to a folder full of images, it doesn't just look for text; it generates a comprehensive semantic understanding of the image.
1. Optical Character Recognition (OCR): It reads all the text in the image (like a receipt or a screenshot of a webpage) and makes it searchable. 2. Visual Description: It understands the objects and context in the photo.
How you use it:
Instead of searching for IMG_4492.jpg, you can search your Mac for:
- *"receipt from the coffee shop"*
- *"screenshot of the error message on github"*
- *"photo of the team at the offsite"*
Dhito instantly returns the correct images because it actually understands what is in them.
Searching Video and Audio with Dhito (Powered by Whisper)
Video files are notoriously difficult to organize. A 2-hour recording of a webinar might contain incredible insights, but finding a specific 30-second clip is a nightmare.
Dhito solves this by integrating OpenAI's Whisper, an incredibly accurate speech recognition model that runs entirely offline on your Mac.
1. Background Transcription: Dhito automatically transcribes the audio of your video files and voice memos in the background. 2. Semantic Indexing: The transcribed text is then fed into Dhito's semantic search engine.
How you use it: You can search for concepts discussed during the video. If you search for *"budget adjustments for Q4"*, Dhito will find the video file and provide a clickable timestamp, taking you exactly to the moment that phrase was spoken. The transcript is indexed at segment level rather than as one block, which is what makes the result a moment rather than a file.
Frames are handled too: as it works through a video, Dhito samples frames and runs the same vision model over them, storing the caption and any on-screen text against that timestamp. So a slide someone shared on a call is findable by what was written on it, not only by what was said about it.
Coverage: 17 video containers — MP4, MOV, MKV, WebM, AVI, M4V, MPEG and others — and 14 audio containers including MP3, WAV, M4A, FLAC, AAC and OPUS.
100% Private and Offline
The most important aspect of searching your personal media is privacy. You don't want your private screenshots, family photos, or confidential company meetings uploaded to a cloud server just to make them searchable.
Because Dhito's vision and audio models run locally on your Mac's hardware, your files never leave your computer.
If you are ready to unlock the information hidden in your images and videos, download Dhito today and experience true multimodal search.
Related Articles
How to Search Inside PDFs on Mac Without Adobe
Looking for an alternative to Adobe Acrobat for searching your PDF collection? Learn the best ways to search inside PDFs on Mac using native tools and local AI.
Mastering Dhito's Search Filters: Finding Files Faster
Semantic search is powerful, but sometimes you know exactly what format or date you are looking for. Learn how to use Dhito's filter tabs to instantly narrow down your AI search results.
How to Build a Local Multimodal Search Engine (Whisper, Florence-2 & MiniLM)
Discover how to combine local AI models like Whisper, Florence-2, and MiniLM to build an offline, multimodal search engine that indexes documents, images, and video transcripts.
Want to try Dhito?
Download Dhito and experience the power of local semantic search today.