Creating paper illustrations

1. Describe the structure first

Click Scientific illustration in a research session, or open it from the toolbox. First describe the components, their relationships, and the reading order. Then specify style, language, and aspect ratio.

Draw a diagram of message passing in a graph neural network.
From left to right, show the input graph, neighbor information aggregation, node representation updates, and graph-level readout.
Use consistent colors for nodes and edges, a white background, and English labels suitable for a research paper.

If you need to reference an existing image, add it as reference material. Before sending, check that the correct session and project are selected.

2. Check and revise each part

First verify the scientific relationships and key terms, then review text, arrows, legends, and layout. Give specific feedback, such as "Change only the aggregation step and keep everything else consistent."

AI images may omit labels or alter the structure. A complete-looking image is not necessarily correct.

3. Save approved results

View generated results in the session files or personal space. Use the available actions to download them, use them in other sessions, or add them to project resources. Adding a result to a project does not move the original session into that project.

For full canvas layout editing, use the standalone drawing editor. It saves editing context under the selected project.

4. Vectorize and export as needed

If an existing image needs a vector format, open Research toolbox → Image vectorization. After conversion, check the text and lines in the SVG, then download the currently available SVG, PDF, or PPTX output.

Follow the target journal or conference's requirements for the final format, resolution, and use of AI images. Keep the source files and local exports for later edits.