A useful revision request separates the content you have approved from the visual changes you want. This tutorial gives a copyable prompt structure for a teaching schematic. It is a planning aid, not a claim that an AI-generated image is scientifically correct.
1. Make a small, source-grounded inventory
Before requesting a revision, write down the objects, exact labels and relationships that the figure is allowed to show. Include a source reference for any scientific statement. If a relationship is uncertain, leave it out of the approved inventory or label the uncertainty explicitly.
Keep visual instructions in a separate list: spacing, alignment, label placement, contrast and the intended reading order. Do not let a request for a cleaner layout become permission to invent additional mechanisms or results.
2. Use a preserve / change / exclude structure
Copy the following template and replace the bracketed fields with your own reviewed information:
Purpose: [The one idea this figure should help the reader understand.]
Approved content: [Objects, exact labels and relationships, with source references in your notes.]
Preserve: [Labels, arrow directions, panel order, quantities and other content that must remain unchanged.]
Change only: [The specific layout or readability changes you want.]
Do not add: [Unsupported objects, mechanisms, numbers, causal arrows, statistical results or decorative details that could be mistaken for evidence.]
After revision: [Items a human reviewer must compare with the approved inventory.]
This is a text template, not executable code or a tested model benchmark.
3. Try a clearly hypothetical teaching example
Imagine a generic three-box diagram labelled Input, Process and Output. Its arrows mean sequence only. No real experiment, biological pathway or dataset is being described.
A layout-only revision request could be:
Keep exactly three boxes labelled Input, Process and Output, in that order. Keep the two existing arrows pointing from Input to Process and from Process to Output. Those arrows mean sequence, not a measured causal effect. Increase spacing between the boxes and place each label inside its box. Do not add feedback loops, extra stages, numbers or new labels. I will compare the revised diagram with this inventory before using it.
The example shows how to constrain a request. It does not show a generated result, and no image-generation credits were used to prepare this tutorial.
4. Compare the revision with the inventory
Do not approve a revision only because it looks more polished. Check the number of objects and arrows, the direction of each connection, every label, and any quantities or annotations. Record unexpected additions or deletions and request a targeted correction.
If the source does not support a new relationship, remove it rather than asking the model to justify it. A plausible-looking explanation is not a replacement for the source or an appropriate subject-matter review.
5. Keep the final handoff explicit
Save the editable source where available, the presentation export and the brief that defines the figure's scope. In the caption, distinguish a schematic explanation from a data plot or a research finding. Check the intended venue's current AI-image and disclosure rules before submission.
In Figran, generated illustrations support text-guided revisions and are raster PNG images. Editable flowcharts and data plots use a separate canvas with SVG export. This tutorial is about how to write and review a revision request; it is not a guarantee that every generated illustration is editable as vectors or acceptable to every journal.
Prepared for Figran with AI assistance. This is an original tutorial from the product's brand, not an independent review, a customer case or validated research. The tutorial text may be reused under CC BY 4.0; that permission does not include Figran's software or brand assets.