Vision Image Indexer
An Obsidian plugin that makes pasted images searchable: the moment an image lands in a note, a vision model writes a structured summary and keywords into a collapsed block beneath it, so screenshots and diagrams stop being invisible to search.
// Images pasted into Obsidian are invisible to search. This plugin has a vision model describe each one, so screenshots and diagrams can be found later.
The brief
Notes fill up with screenshots, slides and diagrams, and every one of them is a black hole: full-text search can’t see inside an image. This plugin listens for the paste event, sends the image to a vision model with a strict structured prompt (title, summary, keywords, and technical fields when they apply), and inserts the response as a collapsed details block right under the image. Visually out of the way, fully indexed by Obsidian search, and richer context for any LLM that later reads the vault.
The engineering
- Catching the paste. Obsidian inserts the image markdown asynchronously, so the plugin polls the cursor line for the inserted link, handling both standard markdown images and Obsidian wikilink embeds.
- Provider portability. One
analyzeImageabstraction dispatches to OpenAI GPT-4 Vision, Google Gemini, or a local Ollama model (llava-llama3), so the cost-versus-privacy trade-off is a settings dropdown, not a fork. - Consistent output. A single hard-coded structured prompt keeps the metadata parseable and uniform across models, with no user prompt engineering required.
Why it exists
It scratches a real itch in my own second-brain workflow, where the vault is also the retrieval corpus for LLM tooling. An image library you can search is a different asset from one you can only scroll.