LogoEdit Banana
GitHub
Powered by SAM 3 & Multimodal LLMs

Edit Banana

Make the Uneditable, Editable

Open Source Core
High Definition Export
Fully Editable

High-Definition Input-Output Comparison

One-to-one comparisons between original static formats and editable reconstruction results. All elements can be individually dragged, styled, and modified.

Flowchart Demo
Static Format
Edit
Original
Editable Reconstruction
Reconstructed
Unexpected Mask
Edit Anything
rewrite your text
Export
Diagram Layout
Static Format
Edit
Original
Editable Reconstruction
Reconstructed
Unexpected Mask
Edit Anything
rewrite your text
Export
Architecture
Static Format
Edit
Original
Editable Reconstruction
Reconstructed
Unexpected Mask
Edit Anything
rewrite your text
Export

Conversion Highlights

Why choose Edit Banana for your content reconstruction?

Layout & Logic

Preserves the layout logic, color matching, and element hierarchy of the original diagram.

1:1 Restoration

1:1 restoration of shape stroke/fill and arrow styles (dashed lines/thickness).

Accurate OCR

Accurate text recognition, supporting direct subsequent editing and format adjustment.

Fully Editable

All elements are independently selectable, supporting native DrawIO template replacement.

How it works

Powered by a robust architecture pipeline

01

Input & Segmentation

Accepts Images/PDFs. Uses fine-tuned SAM 3 mask decoder for precise element segmentation.

02

VLM Scanning & OCR

Fixed Multi-Round VLM scanning (Qwen/GPT-4V) + Azure OCR for text & formulas.

03

Reconstruction

Merges spatial data and text to generate editable XML or PPTX files.