SciFig
Tools & Comparison··Updated

Nano Banana 2.1 vs GPT Image 2.5 for Scientific Figures

Choose Nano Banana 2.1 or GPT Image 2.5 for scientific figures: compare capabilities, adapt research prompts, check labels and continue editing in SciFig.

SciFig Team
SciFig Team
SciFig editorial team

Nano Banana 2.1 and GPT Image 2.5 are both available in SciFig for scientific figure generation and editing.

Start with Nano Banana 2.1 for new labeled layouts; try GPT Image 2.5 Sunburst for focused edits.

A mechanism diagram needs clear relationships; a graphical abstract needs a readable sequence; an existing figure may need just one label corrected. This guide helps you choose a model for that task and turn the brief into a scientific figure in SciFig.

Nano Banana 2.1 vs GPT Image 2.5: Scientific Figure Capabilities

Nano Banana 2.1 updates text and layout; GPT Image 2.5 offers a speed-oriented and a detail-oriented version.

Nano Banana 2.1: Labels and Multi-panel Scientific Images

Google documents improved text rendering, infographic layout and instruction following in Nano Banana 2.1, with 1K, 2K and 4K output. For research illustrations, these updates make it a useful starting choice for named structures, balanced panels and reference-based revisions. It updates Nano Banana 2; Nano Banana Pro remains a separate model in the family.

SciFig: From Research Materials to an Editable Figure

In SciFig, describe your figure or start from a sketch, reference image, PDF or photo. Generate a scientific figure in Illustration. Edit text or icons directly when needed. For element-level editing, you can also vectorize the image to SVG and open it in Canvas. You can also import an existing SVG directly into Canvas.

GPT Image 2.5: Flare or Sunburst

OpenAI positions Flare for fast, high-quality generation and Sunburst for demanding detail and editing precision. Both generate and edit images. In our targeted label-correction example, both Sunburst outputs repaired the wording; both Flare outputs retained a misspelling. The Flare vs Sunburst scientific figure comparison shows the outputs and recorded times.
RequirementNano Banana 2.1GPT Image 2.5
Labeled scientific illustrationsUpdated text rendering and layoutExact-label instructions with Flare or Sunburst
Multi-panel scientific imagesImproved instruction following and infographic layoutExplicit panel order, objects and spacing
Revisions from a referenceConversational image editingFocused editing with preservation instructions
Model choiceUpdated Nano Banana 2 family modelFlare for quick exploration; Sunburst for demanding edits

Which Model Should You Choose for Your Scientific Figure?

Choose by the figure you need to finish: a new composition, exact labels or a targeted correction.

Your taskUseful starting choiceWhat decides the next step
New labeled concept or study overviewNano Banana 2.1Panel arrangement and placement of exact terms
Explore several scientific image layoutsGPT Image 2.5 FlareReading order and how much revision each layout needs
Correct one label in an existing figureGPT Image 2.5 SunburstReplacement spelling and preservation of nearby content
Graphical abstract or mechanism illustrationEither family, with explicit relationshipsArrow direction, object count and visual hierarchy

For a figure dominated by numbers, start from the source table in DataChart. For a process whose boxes and connections need continuous editing, use FlowChart. The image model is most useful for the conceptual illustration within the figure.

See What SciFig Can Do

Choose Illustration for mechanisms and concepts, DataChart for charts from table data, or FlowChart for editable process and structure diagrams.

Explore SciFig

How Do You Write Labels and Symbols in Scientific Images?

Supply the exact terms and characters, then describe where each label and leader line should end.

An eight-label cell illustration needs more than “draw a labeled cell.” List the structures, put the nucleolus inside the nucleus and require each leader line to end on its named structure. That turns a general request into a figure you can inspect term by term.

Example 1
GPT Image 2.5 Sunburst · Output 1
For scientific notation, paste the actual strings: α, β, ΔG, μm, CO₂ and s⁻¹. Our GPT typography outputs retained all six strings in both versions and both runs. Check subscripts, superscripts and line endpoints at the size intended for your article or slide; word presence alone does not establish the biological relationship.

How Do You Structure Mechanisms and Multi-panel Scientific Figures?

Define objects, connections and reading order before asking for color or decorative detail.

A RAF → MEK → ERK illustration can be specified as three boxes, two directed arrows and “Phosphorylation” above each arrow. For a more complex mechanism, write the actual activation and inhibition relationships separately. This makes missing or reversed connections easier to identify.

Example 1
GPT Image 2.5 Flare · Output 1

For a study workflow, assign a purpose to each panel: sample groups, preparation, microscopy and analysis. Our four-panel example uses A–D, five exact labels and three connecting arrows. For a graphical abstract, reduce that sequence to the stages needed to explain the central result, with enough space to distinguish them.

Example 1
GPT Image 2.5 Flare · Output 1

How Do You Edit a Scientific Image From a Reference?

Name one change and its location, then specify the labels, objects and layout that should remain.

For a label correction, write: “Replace ‘Treatmet group’ with ‘Treatment group’ at the top right of panel A; keep the same font, baseline and color.” Then name the other panels, labels and arrows to preserve. For a reference-based scientific figure, say whether the input supplies the composition, objects or visual style.

The two-version editing example displays both outputs from each version. Use it to inspect the corrected label and the surrounding scientific figure, rather than judging the replacement in isolation.

A Short Prompt Structure for Scientific Figures

A useful scientific figure prompt specifies content, layout, relationships, exact labels and visual style.

Use this structure with either model, replacing the bracketed fields with your research content:

Create a [figure type] showing [objects]. Arrange them in [layout]. Connect [source] to [target] with [relationship]. Use exactly these labels: [label list]. Keep [style and spacing].

For an edit, add the target change and the elements to preserve. Keep the first attempt focused on the explanation; adjust one failed requirement at a time. Select an example image above to open the complete prompt, copy it or generate in SciFig. The scientific AI prompting guide develops the same method for more complex figures.

Tip

Before generating, write a short checklist: required objects, label spelling, arrow directions and panel order. Use that checklist to decide what to revise.

Start Creating Scientific Figures

Choose Illustration for mechanisms and concepts, DataChart for charts from table data, or FlowChart for editable process and structure diagrams.

Start Free

Create, Edit and Export Scientific Figures in SciFig

Choose the workspace by the output: conceptual image, data plot or editable process diagram.

TaskSciFig workflowEditable output
Concept or mechanism illustrationIllustration from text or reference materialsEdit text and icons directly; optional SVG vectorization for element editing
Chart from numerical resultsDataChart from your source tableRefine the chart and convert it to SVG
Process or structure diagramFlowChart from stages and relationshipsNative SVG
Start your next figure in SciFig. Use Canvas to adjust recovered vector elements or continue editing an existing SVG. For a visual explanation spanning several panels, the graphical abstract guide helps you organize the scientific message first.

Scientific Figure FAQ

Keep going

Recommended resources

Templates, tools, and further reading for your research

Call to action background

Ready to start?

Publication-ready scientific figures, in minutes

Start Creating Free

Free to start · No credit card required · Built for researchers