Practical checks for the brief, the first draft and the exported file. Your research stays the source of the scientific content.
Write a drawing brief the model can follow
Start with the figure’s job: explain one structure, compare supplied states, show a documented relationship, or summarize a study. A list of attractive objects rarely tells the reader what the figure means.
Name the structures that must appear and the exact labels you want. State relationships separately from visual preferences. Use a spatial description for layout and explicit wording for causal or directional relationships.
Supply the components, their names and the relationships supported by your source.
Say which details should remain unspecified. Do not ask the model to infer missing results.
Choose the reading order: left to right, overview to detail, or a central concept with supporting views.
Keep colors and visual style in a separate instruction so they do not change the science.
Decide whether a reference supplies scientific content, composition, color, or visual style. Tell the model which role it plays. A beautiful reference can still show a different organism, scale or mechanism from your study.
Use material you may share. A reference uploaded for generation leaves the browser and is handled by the configured image service. Some services use temporary publicly accessible links, so confidential or unpublished material needs particular care.
Describe what to retain and what should change.
Check names, orientation and scale against your own source.
Remove personal identifiers, private filenames and irrelevant research details.
Review the upload disclosure before sending the file.
Treat the first result as a conceptual draft. Plausible-looking structures and correctly spelled labels do not establish scientific accuracy. Review the entire figure against the source material, including areas you did not ask to change.
For a mechanism, check each arrow individually. For a comparison, make sure both views use a consistent scale and visual treatment. For a graphical abstract, verify that the claimed message is actually supported by the study.
Confirm structures, orientation, component counts and terminology.
Read every label at the size it will appear in the manuscript.
Check arrow direction, implied causality and missing intermediates.
Keep measured results and experimental images separate from generated conceptual artwork.
After an AI edit, inspect both the requested region and surrounding content.
The AI illustration is a bitmap. Text labels and arrows added in ResearchFig are separate objects. You can move and edit those annotations, and keep them separate when exporting SVG or PowerPoint.
A guided image edit makes a new bitmap. Painting a mask expresses where you want a change; it is not a guarantee that every other pixel will remain identical. Compare the proposed image before accepting it.
Use separate annotations for wording you expect to revise.
Keep a saved version before a meaningful image edit.
Open the exported SVG or PPTX and check the annotations in the target application.
Do not describe an embedded bitmap as fully vector artwork.
Start with the journal’s current figure instructions. Pixel dimensions, print width, text size and file format are separate requirements. Changing a DPI metadata field does not add missing image detail.
PNG and JPG provide raster output. SVG and PPTX preserve the added annotation objects while embedding the generated image. Inspect the exported file itself rather than relying only on the browser preview.
Check actual pixel dimensions and the final print width.
Inspect label legibility, contrast and line weights at manuscript size.
Use the journal’s own current instructions for resolution and format targets.
Keep a private editable project or version for later corrections.
The local bar graph, histogram, box plot and heatmap tools render values you provide. They do not ask an image model to invent a chart. The included datasets are synthetic interface examples, not research results.
Check the chart method as well as its appearance. Histogram bin boundaries, box-plot quartiles and heatmap color ranges can change how a reader interprets the same data.
Replace all example values with your own verified data.
Review units, axis ranges and group labels.
For box plots, ResearchFig documents R-7 quartiles and Tukey whiskers.
Export SVG when you need actual vector chart elements.