- 1Add a file
- 2Make cards
- 3Get your deck
Your cards
Diagrams
Click a mask to hide or show that label. Drag to move, drag the corner to resize, drag on the image to draw a new one, Delete to remove.
Tested on 16 labelled diagrams
Before every change to label detection we run the same diagrams through it and count, for each label, whether it ends up under exactly one mask that is switched on — nothing for you to click. These are the results from 17 October 2026, with the text-detection engine the site uses now.
| Diagram | Labels masked, no clicks |
|---|---|
| Heart, textbook drawing, two-line labels | 14 / 14 |
| The same heart, photographed off a screen (tilted, blurred, JPEG) | 14 / 14 |
| Brain, inferior view, cranial nerves on label tags | 12 / 12 |
| Animal cell, labels stacked one line apart | 21 / 21 |
| Nephron, small image with three-line labels | 7 / 7 |
| Skeleton, tall figure with group headings | 26 / 26 |
| Limbic system, rendered illustration with a title and headings | 10 / 10 |
| Respiratory tract, repeated names (“Superior lobe” twice) | 46 / 48 |
| Skull, numbers 1–11 only | 11 / 11 |
| Tooth, numbered figure with a key | 10 / 10 |
| Lecture-slide screenshot with a 1909 plate of the sacrum | 4 / 4 |
| Eye, phone photo of a printed page with a hand shadow | 18 / 19 |
| Brachial plexus schematic, lower-case nerve names | 23 / 23 |
| Brain lobes, white labels printed on colour | 4 / 4 |
| Eye, Chinese labels | 18 / 19 |
| Gray’s Anatomy plate (1918), tiny abbreviations like “L.C.C.” | 5 / 11 |
| All 16 | 243 / 253 (96%) · median wait 10 s |
What went wrong: most of the tiny abbreviations on the century-old engraving were not read at all; two neighbouring labels on the respiratory figure were merged into one mask; two three-line labels were split in two. Each is a click or a drag in the editor. Most of these diagrams were also used to tune the masking rules, so read the numbers as a best case for printed labels.
Built for the way Anki does occlusion now
- The built-in note type. Since 23.10 Anki has its own Image Occlusion note type; decks made for the old add-on often show no masks. Ours are the built-in kind, so Anki's own editor can open and change them.
- Every label, one card each. Choose “hide all, guess one” or “hide one, guess one”. Merged multi-line labels (“Superior vena cava”) become one mask.
- Captions stay visible. Long captions and big titles start unmasked. Turn any mask on or off with a click.
- From your PDF too. Add a lecture PDF and press “Occlude” on any page to turn that page into an occlusion note.
How to make image occlusion cards from a diagram
- Add the diagram. A PNG or JPG, a screenshot pasted with Ctrl+V (⌘V on a Mac), or a page of a PDF via “Occlude”. One credit per diagram.
- Check the masks. Every label we found is already covered. Click a mask to hide or show that label, drag to move or resize, draw a new one where a label was missed.
- Pick a study style and download. “Hide all, guess one” or “hide one, guess one”, then one .apkg. Each mask that is on becomes its own card.
Image occlusion in Anki, explained
The built-in note type and the old add-on
Before version 23.10, image occlusion in Anki meant installing the Image Occlusion Enhanced add-on, which stored its own mask images for every card. Anki 23.10 made occlusion part of Anki itself: one note holds the original picture and a list of mask shapes, and Anki draws the masks when it shows the card. AnkiMobile and AnkiDroid read the same note type, so a diagram you make on a laptop studies the same way on a phone.
Many decks shared online are still the add-on kind, and opened in a recent Anki without the add-on they can show the picture with no masks at all. AnkiForge writes the built-in kind directly — the note type carries the same markers Anki’s own editor gives it, so Anki treats the notes as image occlusion. We check that on every change by importing a sample deck with Anki’s official library.
“Hide all, guess one” or “hide one, guess one”?
Hide all, guess one covers every label and highlights the one being asked, so neighbouring labels can’t give the answer away. It suits structures you have to place from scratch, like the cranial nerve exits on the base of the brain. Hide one, guess one covers only the label being asked and leaves the rest readable — gentler, and better for dense figures where the surrounding names are the point, like the organelles of a cell. It is the same switch Anki shows in its own occlusion editor.
Diagrams that work well, and ones that don’t
Printed, horizontal labels of a few words are the easy case, and labels that wrap onto two or three lines are joined into one mask. Long captions, titles and section headings start unmasked, so they don’t become cards nobody wants. The hard cases are handwriting, very small text, abbreviations on old engravings and labels printed over busy textures — that is where labels go missing in our tests. Figures numbered 1, 2, 3 with a separate key are read too, but a card that hides “3” asks for a number, not a name; masking the names in the key usually makes better cards.
Changing masks after import
The notes stay ordinary Anki image-occlusion notes. Open one in Anki’s Browse window and you get Anki’s occlusion editor, where you can add, move or delete masks; the cards follow the next time they come up.
Pay per page. Start free.
30 free credits when you sign up, then packs from $4.99. 1 credit = 1 page or slide (or a notes section of up to ~3,000 characters) turned into cards, or 1 diagram scanned for labels. Failed runs are refunded and downloading is free. See all credit packs
FAQ
How accurate is the label detection?
On our test set of 16 labelled diagrams, 96% of labels ended up under exactly one mask with nothing to click. Handwriting, very small text and labels drawn over busy images can be missed or split — that is why every mask stays editable before export, and you can draw extra masks by hand.
What is “hide all, guess one”?
Every label is covered and one is highlighted as the question. “Hide one, guess one” covers only the label being asked. It is the same choice Anki offers in its own image-occlusion editor.
Which images can I use?
PNG, JPG, WebP and most other images your browser can open, or a screenshot pasted straight into the page. Large images are scaled to at most 2,400 pixels on the long side before scanning, and that same image goes into your deck so the masks line up.
Which Anki versions can open the deck?
Cloze cards work in any recent Anki, AnkiMobile and AnkiDroid. Image-occlusion notes use the built-in Image Occlusion note type, which needs Anki 23.10 or newer on desktop and a current AnkiMobile or AnkiDroid.
How much does it cost?
New accounts get 30 free credits. After that, credit packs start at $4.99. One credit is one page or slide turned into cards, or one diagram scanned for labels. If a run fails, its credits are refunded automatically. Downloading is free.