figures.science
From your inputs to a usable figure
Choose precise drawing or AI illustration, understand revisions, and see which parts remain editable in each output.
Choose a route
Use a precise tool when numbers, counts, labels, and relationships determine the figure. Choose an illustration when the task calls for a visual scene. Graphical Abstract and CONSORT offer precise and Illustration routes on one page with separate outputs. Medical, Scientific Image, Schematic and Biology are independent raster workspaces. CONSORT Illustration does not check count arithmetic.
Check the source material
Review missing fields, conflicting counts, event codes, and relationships. Scientific and statistical checks belong to the individual tool; a plausible image cannot replace them. Dot Plot validates values, PRISMA checks stage counts, and Framework checks supplied directions and constructs before drawing.
Revise and export
Twelve native tools rebuild from reviewed fields and connect SVG/PPTX, PNG and basic-Latin vector PDF; data plots also export source CSV and Palette has five code formats. Sentence roles require explicit review. The AI illustration routes produce independent raster PNGs and accept one eligible PNG reference for image revision; exact anatomy, words and pixels are not guaranteed. Reviewed fixed examples use self-authored simulated inputs and do not guarantee every possible result.
Save and return
Guest native edits stay in the current tab unless you explicitly save a confirmed figure. Saved native works restore the Figure data, available source CSV and exports, rather than the original domain form. Saved AI works point to your owned image task; version history keeps earlier saved revisions available.
The figure assistant
All 22 tools connect AI input organization into unconfirmed fields. Review and apply those fields, then create the native figure or explicitly generate the raster illustration; the provider’s reported usage is shown. The assistant is instructed to preserve supplied facts and leave unknown fields blank; its output still needs review and domain validation. Input organization does not use image credits and has separate rate limits. It does not execute model-generated code. A combined image-and-label editing layer belongs to version two.