AI box extraction

When creating a diagram card, drawing every box by hand over a complex image can take time. Automatic AI extraction analyzes the image and proposes boxes over identifiable labels and elements, so you can start from a solid base rather than from scratch.

It's an assistant, not an oracle. The result is a proposal, not a final answer. Always review the generated boxes before saving the card.

What it does exactly

The AI examines the image you uploaded and detects areas that look like labels, captions, or distinct diagram elements. For each area it finds, it proposes a rectangular box with its position and size. The boxes arrive numbered, ready to use in review.

The process does not modify the image; it only adds boxes on top of it, exactly as if you had drawn them by hand.

How to run the extraction

1. Open or create a New Diagram Card.
2. Load the diagram image (drag it in or paste with Ctrl+V).
3. Once the image is loaded and the box editor appears, find the Extract boxes with AI button in the editor's top bar.
4. Press the button. While the AI works, the button shows a loading indicator.
5. When it finishes, the proposed boxes appear on the image, numbered in order.

Extraction uses AI quota from your plan. If you run out of available tokens, a notice appears with an option to upgrade. See plans and limits for more information.

What you get

You receive a set of boxes drawn on the image, positioned based on what the AI detected. The number of boxes varies with image complexity; there is no fixed value.

You can see how many boxes were created in the status message below the image.

Reviewing, adjusting, and deleting the proposed boxes

The AI result is always a starting point. After extraction you can:

• Delete an individual box — click directly on the numbered box. It disappears immediately. Do this for any box covering an incorrect or uninteresting area.
• Draw additional boxes — press Enable Drawing to enter drawing mode and add boxes the AI missed.
• Clear everything and start over — press Clear All to remove all boxes and draw from scratch. You can run extraction again if you like.
• Undo the last drawn box — press Ctrl+Z (or Cmd+Z on Mac) while the modal is open.

Once you're satisfied with the result, press Create (or Save if editing) to save the card.

Which images work best

Extraction gives the best results when:

• The diagram has clear text labels next to its elements (part names, symbols, abbreviations).
• Contrast between the background and the elements is high.
• The image has good resolution and is not heavily compressed.
• Labels do not overlap each other.

It works less well with very dense images without text, unannotated photographs, or diagrams with a highly detailed background. In those cases it can still provide a useful first draft, but it will need more manual adjustment.

Frequently asked questions

Can I run extraction more than once on the same image? Yes. Every time you press Extract boxes with AI, the previous result is replaced by the new one. Each attempt uses AI quota.

Does the AI know what each box means? No. The AI detects areas of the image that look like labels or elements, but it does not generate answer text for each box. You learn the content of each zone during review.

What if the AI detects no boxes? You can draw boxes by hand with Enable Drawing. The image is already loaded, so you just need to mark the zones that interest you.

Does extraction work with screenshots and photos of books? Yes, as long as the image has enough resolution and contrast. A dark or blurry photo will give worse results.

Next step

• Diagram cards — how to create and use diagram cards from scratch.
• Plans and limits — how much AI quota your plan includes.
• AI cards — other AI features available in EON for text cards.