by SciFigure Pro (LX AI)
SciFigure Pro helps researchers prepare publication-oriented figures — a deterministic quality checklist and caption guidance so you catch dpi, legend, and honesty issues before submission.
Decision-support only — confirm before you publish or act.
I built SciFigure Pro around a set of hard rules that run locally in the browser. When you drop a figure file (PNG, TIFF, PDF, or SVG) onto the upload zone, the tool reads the image metadata and pixel data without sending anything to a server. The first rule checks the file format against a whitelist of accepted types; if the format is not on the list, the tool returns a clear message asking you to convert the file.
The ingestion step extracts embedded ICC profiles, bits per channel, and color space information. For TIFF files it also reads the XResolution and YResolution tags. For PNG it looks at the pHYs chunk. If any of these tags are missing, the tool flags them as “missing metadata” and suggests adding them in your image editor.
After metadata is gathered, the engine runs a series of Boolean tests. Each test returns true or false and is linked to a specific guideline from major publishers (Elsevier, Springer, Wiley, PLOS). The tests are independent, so a failure in one area does not affect the score of another. The output is a simple table that lists each rule, its status, and a short remediation tip.
Because the rules are deterministic, you can reproduce the same result on any machine with the same browser version. This makes the checklist suitable for lab notebooks and for sharing with co‑authors who need to verify compliance before a group meeting.
Once the checklist passes, SciFigure Pro moves to caption assistance. The tool does not write the caption for you; instead it offers a structured template based on the figure type you select (micrograph, graph, schematic, or photograph). You fill in the placeholders and the tool checks that each required element is present.
When you click “Add caption”, a modal shows four icons. Choosing an icon loads a corresponding template:
If you try to advance without filling a mandatory placeholder, the tool highlights the empty field in red and displays a tooltip with the expected format.
After you type the caption, the tool runs three quick checks:
These checks are purely advisory; you can override them if you have a justified reason, but the tool logs the override for your own records.
DPI (dots per inch) is a frequent cause of rejection because many authors assume that a figure that looks sharp on screen meets print requirements. SciFigure Pro calculates the effective DPI from pixel dimensions and physical size, then compares it to journal‑specific thresholds.
If the file contains explicit resolution tags (TIFF XResolution/YResolution, PNG pHYs), the tool uses those values directly. When tags are absent, it falls back to a heuristic: it assumes the intended print width is the width you specify in the “Target width” field (default 80 mm for a single column). The formula is:
DPI = (pixel width / target width in mm) × 25.4
The same calculation is performed for height; the lower of the two values is reported as the effective DPI.
| Figure type | Minimum DPI (print) | Minimum DPI (screen‑only) |
|---|---|---|
| Line art / vector | 300 | 150 |
| Halftone photograph | 300 | 150 |
| Micrograph (fluorescence) | 600 | 300 |
| Schematic with fine lines | 600 | 300 |
If the calculated DPI falls below the applicable minimum, the tool shows a red warning and suggests two actions: increase the pixel dimensions by rescaling in your editor, or reduce the target print width. The tool does not resample the image for you because that could introduce interpolation artifacts; it only advises.
Legends are often overlooked, yet they are a major source of confusion for readers. SciFigure Pro extracts text objects from SVG and PDF files (using client‑side parsing) and compares them against the legend entries you provide in the caption.
For SVG, the tool reads all <text> elements and records their string content, font size, and fill color. For PDF, it uses the built‑in PDF.js library to draw the page to a canvas and then attempts to recover text via the hidden text layer; if no text layer exists, it falls back to optical character recognition via Tesseract.js (compiled to WebAssembly) – a slow but reliable fallback.
You paste the legend items as a comma‑separated list in a small field. The tool normalizes each item by trimming whitespace, converting to lower case, and removing punctuation. It then attempts to match each normalized item to any extracted text object using a Levenshtein distance threshold of 2. Items that do not find a match are highlighted as “missing in figure”. Conversely, any extracted text object that does not match a legend item is flagged as “extra in figure”.
Beyond presence, the tool verifies that font size and color are consistent across all legend entries. If you have used two different font sizes, the tool warns that the legend may appear uneven. It also checks that the legend background contrasts sufficiently with the figure background (WCAG AA contrast ratio ≥ 4.5:1).
Scientific integrity depends on transparent reporting of how a figure was generated. SciFigure Pro includes a set of honesty checks that remind you to disclose critical details. These checks are not guarantees; they are prompts to verify your own records.
When you indicate that a figure was created with a specific software (e.g., ImageJ, Prism, Illustrator), the tool asks you to include the version number. If you omit the version, a yellow banner appears: “Version missing – consider adding it to the caption or methods”.
For graphs generated from raw data, the tool encourages you to list key parameters: bin width, smoothing threshold, regression model, and random seed if applicable. You can enter these parameters in a collapsible box; the tool does not validate their correctness, but it checks that the box is not left empty when you have selected “Graph” as figure type.
If your figure is based on a dataset, the tool prompts you to add a data availability statement (e.g., “Raw data are available at https://doi.org/10.xxxx/xxxxx”). The statement is not checked for existence; it is merely a reminder to include it somewhere in the manuscript.
To align with emerging best practices for AI‑assisted figure generation, I have incorporated references to two public frameworks. The first outlines safety considerations for large language model applications (https://owasp.org/www-project-top-10-for-large-language-model-applications). The second provides a structured approach to managing AI risk (https://www.nist.gov/itl/ai-risk-management-framework). Mentioning these does not imply that SciFigure Pro uses AI; it simply offers resources for those who do.
These flags are decision‑support only. They cannot detect intentional misrepresentation, nor can they guarantee that all relevant details have been disclosed. You remain responsible for verifying that your figure legend, methods section, and data statements are complete and accurate.
Researchers often move between reference managers, LaTeX editors, and word processors. SciFigure Pro is designed to fit into that workflow without requiring you to change your preferred tools.
After you finish the checklist and caption, you can download a small JSON side‑car file that contains:
You can place this JSON file alongside your manuscript source. A companion script (provided on the GitHub repo) reads the JSON and inserts a ready‑to‑print figure label into LaTeX or Word via a simple macro.
If you prefer to work directly in your editor, the tool offers a “Copy caption” button that places the formatted caption on your clipboard. The caption includes line breaks that match the target journal’s style (you select the style from a dropdown: Elsevier, Springer, Nature, or Custom).
Because all processing happens client‑side, the original figure file is never altered. The JSON side‑car is plain text and can be tracked by Git. This lets you review exactly what checks were run at the time of figure creation, which is useful for reproducibility audits.
The tool is a single‑page application that loads once from the server and then runs entirely in the browser. After the initial load you can disconnect from the internet and continue to work on figures; the only server interaction is optional anonymous usage statistics that you can disable in the settings.
SciFigure Pro is not a substitute for careful reading of the author guidelines. It does not guarantee that your figure will pass editorial review, nor does it promise 100 % detection of every possible issue. The tool is decision‑support only; you must still verify the output yourself.
It is not an image editor. You cannot use it to resize, recolor, or retouch photographs. If you need to adjust brightness, contrast, or remove background noise, you must do that in a dedicated editor before uploading.
It is not a plagiarism detector. The tool does not compare your figure against a database of published images to check for duplication.
It is not a statistical validator. It does not assess whether the data underlying a graph are appropriate or whether the statistical tests are correctly applied.
It is not a replacement for a co‑author’s review. Sharing the JSON side‑car with a colleague can help, but you should still discuss the figure in a group meeting.
Finally, it is not a cloud‑based service that stores your files. All processing stays on your machine unless you explicitly choose to upload the side‑car for backup.
SciFigure Pro helps researchers prepare publication-oriented figures — a deterministic quality checklist and caption guidance so you catch dpi, legend, and honesty issues before submission.
Decision-support only — confirm before you publish or act.
Want to try the product? SciFigure Pro
Generated 2026-09-17 · primary sources only · no fabricated traffic metrics