Draw a neural network diagram with AI
Turn a PyTorch model definition into an editable Excalidraw scene, with a block overview or a closer view of individual operations.
To draw a neural network diagram with AI, give the agent the Python file that defines the network. The nn2excalidraw tool reads its structure and lays out an editable scene. A block view makes a useful slide for a lab meeting, while a more detailed view can help a colleague follow the operations inside a stage.
If you know the real input shape, include it. Otherwise the diagram can still show properties provided by the constructor. Static analysis reads the source without importing it. A separate tracing mode executes the supplied model and is suitable only for code you trust.
What it does
- Show where a residual path rejoins the main stream.
- Keep repeated blocks together for a compact overview.
- Expand the layout when you need to discuss individual layers.
- Change the drawing direction to suit the page.
- Export a drawing for an Obsidian vault with the Excalidraw plugin.
Install and use
- Install the skill for your agent and the maker’s Python command line package. Static mode needs no GPU, trained weights or PyTorch installation.
- Ask, “Read this file in static mode and draw the main network class with its skip connections.” Name the class if the file defines several candidates.
- Open the resulting scene in Excalidraw. Check the reported entry class and connection counts against the model before using it in a report.
FAQ
Source
Made by bot66. Original skill: nn2excalidraw, published on GitHub under the MIT licence.
View on GitHub