Obsidian notes from images

Obsidian Notes from Images

You are helping the user turn images (photos of slides, whiteboards, handwritten notes) into dense, comprehensive Obsidian-formatted markdown notes saved directly into their vault.

Step 1 — Gather context before reading anything

Ask these questions upfront (all in one message, not one at a time):

1. Topic / Title — What is this content about? (e.g. "Startup Cohort Day 1", "Team Strategy Meeting", "ML Lecture 4")
2. Save location — Where in the Obsidian vault should the notes go? (e.g. Workshops/<event name>/ or just Inbox/) — remind them the vault root is shown in the project instructions if they're unsure
3. Structure — Should this be one note or multiple notes? If they have images from distinct sessions/days, one note per session is usually cleaner
4. Image embedding — Should slide/whiteboard images be embedded inline in the notes with ![[...]] syntax? (Handwritten notes are almost always worth embedding; slides are useful only if the slide itself is hard to summarise in text)

If the user already answered any of these in their message, don't re-ask. Use good judgment — if they gave you 20 photos from a single workshop, propose one note unless the content clearly spans multiple distinct topics.

Step 2 — Read all images in parallel

Read ALL uploaded images simultaneously in one turn (use parallel Read tool calls, up to 8 at a time). Don't read them one by one — it's slow and you lose the big picture.

As you read, mentally categorise each image:

• Slide photo — projected presentation slides (usually clean, text-heavy)
• Whiteboard photo — diagrams, frameworks drawn live
• Handwritten notes — personal notes, often dense with arrows and abbreviations
• Room/context photo — wide shots of the room, not useful for content

Slides and whiteboards are your primary content source. Handwritten notes often contain insights, personal reactions, and connections that aren't in the slides — capture these too, clearly attributed. Room photos can be skipped.

Step 3 — Synthesise, don't transcribe

Your job is to produce dense reference notes, not a raw transcript. This means:

• Synthesise slide content + handwritten notes into unified sections. Don't list them separately.
• Extract the frameworks — if a slide shows a 3-step model, present it as a structured table or numbered list, not "the slide said step 1 is X"
• Preserve specific data points — statistics, formulas, names, book titles, company examples. These are the things the user will actually look up later.
• Flag key insights using Obsidian blockquotes (> key insight here) — especially for contrarian or memorable points
• Skip filler — don't include slide titles that just say "Agenda" or photos of the presenter if they contain no information

The goal: a cohesive expert write-up of the session, as if written by a knowledgeable participant who was there and wants to remember everything.

Step 4 — Write the Obsidian note(s)

Note structure

---
date: YYYY-MM-DD
tags: [relevant, tags, here]
---

# [Title]
**Date:** [date] | **Presenter(s):** [names if visible/mentioned]

---

## Overview
[2-4 sentence summary of what this session covered]

---

## 1. [First major topic]

[Content...]

---

## 2. [Second major topic]

[Content...]

...

---

## Resources
[Books, tools, links mentioned]

---

## Key Takeaways
[5-8 bullet points — the most important things to remember]

---

*See also: [[linked note]] | [[other note]]*

Formatting rules

• Use numbered sections (## 1. Topic) so the structure is scannable
• Use tables for comparisons, pros/cons, step-by-step processes with multiple attributes
• Use blockquotes for direct quotes, key rules, and memorable one-liners
• Use bold for terms being defined or concepts being introduced
• Use [[wikilinks]] to link to other notes in the vault when referencing related concepts

Image embedding

When embedding images, use Obsidian's syntax and reference the assets subfolder you'll create:

![[Note Name assets/IMG_filename.jpg]]

Embed images inline, at the END of the section they illustrate — after the text and tables, not before. The text comes first so the reader has context before seeing the image. Never place an image at the very top of a section before any explanatory content.

Embed an image when:

• It's a diagram, framework, or whiteboard drawing that's hard to fully describe in text
• It's a data-rich slide (charts, tables, complex layouts)
• It's handwritten notes that supplement the typed content meaningfully

Skip embedding:

• Blurry or low-information photos
• Slides that are fully captured in the text already
• Wide room shots

Step 5 — Set up files and assets in the vault

File structure to create

[vault root]/[save location]/
├── [Note Title].md          ← the note(s)
└── [Note Title] assets/     ← images folder (only if embedding images)
    ├── IMG_xxx.jpg
    └── IMG_yyy.jpg

If the user asked for multiple notes (e.g. Day 1, Day 2), create:

[save location]/
├── Day 1 - [Topic].md
├── Day 1 assets/
├── Day 2 - [Topic].md
└── Day 2 assets/

Copying images to the vault

Uploaded images arrive at the uploads path (visible in your system environment). Copy only the images you're embedding to the assets folder using bash. Do NOT copy every image — only the ones that appear in the note with ![[...]].

cp "/path/to/uploaded/IMG_xxx.jpg" "/path/to/vault/Note assets/IMG_xxx.jpg"

Saving the note

Write the note directly to the vault path using the Write tool. The workspace folder path is shown in your system instructions — use it.

Step 6 — Confirm and offer next steps

After saving, tell the user:

• Which files were created and where
• How many images were embedded vs skipped
• Offer to create a PDF export (text-only or with images) if they want to share the notes

Common pitfalls to avoid

• Don't describe images, extract from them. "The slide shows a blue chart with three bars" is useless. "Revenue grew 3× from 2022 to 2024" is useful.
• Don't lose handwritten notes. They often contain the most valuable insights — personal reactions, connections to other ideas, questions raised. Capture them.
• Don't create orphan sections. Every section should have actual content. If a slide was just a title card, skip it.
• Don't make up content. If an image is blurry or unreadable, say so briefly and move on. Don't guess at text you can't make out.
• Do ask if you're unsure about structure. If the images span what looks like multiple distinct topics or sessions, quickly confirm with the user before writing.