Source retained
The original stays available beside later drafts.
AI-assisted revision for scientific schematics
Revise scientific figures with AI.
Upload a scientific schematic, mark what needs changing, and describe your revision. Generate a new version while keeping the original available for comparison.
Start from the figure you already have.
Upload a scientific schematic, choose an example, or paste a revision request. Each result starts a new version in your conversation.
Change CNN to Visual Encoder, make the arrows teal, and align the three modules.
Choose a guided revision
A controlled scientific figure revision workflow
Region guidance
Use marks and notes to point at the exact change.
Version continuity
Continue from a chosen result without losing context.
Human verification
Every generated relationship remains an author decision.
Guided revision cases
See the source, instruction, and revision together.
Each case shows a narrow revision request rather than a polished output with no context.
Methodology workflow
Clarify a crowded research pipeline
The source uses uneven spacing and a generic central label. The revision creates a clearer sequence without changing the stated input and output.
Revision prompt
Change CNN to Visual Encoder, align the three modules, and use teal arrows. Preserve the input and output labels.
- Aligned the processing modules on one baseline
- Replaced the requested component label
- Applied one directional color to the workflow arrows
Multimodal AI architecture
Separate modality encoding from cross-modal fusion
The source mixes image and text pathways with direct task connections. The revision introduces explicit stages, a shared representation, and a distinct training-only objective.
Revision prompt
Separate MRI and clinical-text inputs into parallel encoder lanes. Route both into cross-modal attention, then a shared embedding and two task heads. Keep the auxiliary alignment objective as a dashed training-only path.
- Separated image and language encoders into parallel lanes
- Distinguished inference flow from the training-only objective
- Preserved both prediction heads and the shared representation
Longitudinal study design
Clarify cohort allocation and biospecimen timing
The revision separates participant flow from the assessment schedule so readers can trace cohort allocation without losing collection time points.
Revision prompt
Place screening, eligibility, allocation, and the primary endpoint in one participant-flow row. Show Cohort A and Cohort B as separate branches. Move MRI and blood collection to a labeled visit timeline and preserve every source time point.
- Separated participant flow from scheduled assessments
- Made cohort convergence at the primary endpoint explicit
- Preserved all visit and biospecimen time points
Cell signaling pathway
Make compartment boundaries and feedback readable
The revised pathway assigns each entity to its biological compartment, separates activation from inhibition, and explains the notation in a compact legend.
Revision prompt
Organize Receptor R, Kinase A, Kinase B, TF-X, Inhibitor I, and target genes by extracellular, cytosolic, and nuclear compartments. Use solid teal arrows for activation and dashed red blunt-ended lines for inhibition. Preserve every named relationship.
- Placed each entity in the correct labeled compartment
- Used distinct notation for activation and inhibition
- Retained the feedback loop and all named relationships
These are guided product examples, not customer results. AI revisions can change scientific meaning and require human review.
Open the case libraryRevision capabilities
Six visual changes, one traceable workflow.
A useful editor must understand the requested change and the information that cannot move with it.
Correct labels
Rename a module or shorten a caption without rewriting neighboring terminology.
Replace “CNN” with “Visual Encoder.” Keep every other label unchanged.
Reroute connections
Untangle arrows, distinguish activation from inhibition, and preserve directionality.
Separate the feedback path from the main flow. Preserve all source and target nodes.
Rebuild hierarchy
Group modules by stage, compartment, cohort, or experimental sequence.
Group the blocks into acquisition, encoding, fusion, and prediction stages.
Normalize visual encoding
Use color, stroke, shape, and spacing consistently across one figure.
Use teal for inference, amber for training-only paths, and add a compact legend.
Recompose panels
Clarify panel order and align repeated elements across a multi-part figure.
Place panels A and B on the first row. Align captions and keep panel lettering fixed.
Reduce diagram density
Improve scanning by separating crowded regions without deleting scientific content.
Increase spacing around the pathway branch. Do not remove any named entity or edge.
Different figures need different revision logic.
A pathway, study design, model architecture, and graphical abstract each have their own structural details to protect.
Revision focus
Make compartment boundaries and feedback readable
Clarify compartment boundaries and causal direction while retaining every named biological entity and relationship.
Author checkpoints
- Entity placement matches its biological compartment
- Activation and inhibition use distinct notation
- Feedback direction agrees with the source
Organize Receptor R, Kinase A, Kinase B, TF-X, Inhibitor I, and target genes by extracellular, cytosolic, and nuclear compartments. Use solid teal arrows for activation and dashed red blunt-ended lines for inhibition. Preserve every named relationship.Inspect the full case
Cell signaling pathwayRevised schematic
A revision loop you can inspect.
Keep the source visible, make the requested change explicit, and review the output before it leaves your workspace.
Upload the source
Begin with the scientific schematic you need to revise.
Mark the target
Use an annotation guide when a change applies to one region.
Describe the revision
State what must change and what must remain untouched.
Compare and verify
Review labels, relationships, and scientific meaning against the source.
Prompt structure
Write revision prompts like an author.
The strongest instruction defines the requested change, the protected content, and the details that need checking.
Example revision prompt
Separate the MRI and clinical-text inputs into parallel encoder lanes. Preserve both prediction heads and the auxiliary alignment objective. Verify that the dashed path is labeled training only and that no source relationship is removed.
Change
Name the exact label, connection, region, hierarchy, or visual encoding that should be revised.
Preserve
State which entities, directions, terms, panel letters, and relationships must remain untouched.
Verify
Identify the scientific details the author must compare with the source after generation.
Specific constraints reduce ambiguity, but every generated label and relationship still requires human review.
Reviewer and PI feedback
Translate a comment into a concrete revision plan.
Paste the feedback beside the figure. The assistant can turn it into explicit visual changes before generating a new version.
Reviewer comment
“Figure 2 is difficult to follow. Please clarify the relationship between the encoder and decoder modules and improve label readability.”
Revision plan
- Separate encoder and decoder stages
- Redraw the main directional connection
- Increase label size without changing terminology
The plan and generated figure still require author verification.
Revision record
Figure 2 · methodology workflow
Instruction used
Change CNN to Visual Encoder, align the three modules, and use teal arrows. Preserve the input and output labels.
Labels checkedDirection checkedSource retained
Keep the reasoning behind every figure change.
A revision is easier to approve when the source, instruction, result, and author checks remain connected in one record.
Compare instead of overwrite
Inspect the new draft against the exact source that produced it.
Continue from an approved direction
Use a selected result as the basis for the next narrow revision.
Retain the revision rationale
Keep the instruction beside the result so collaborators can understand why it changed.
Human verification
A professional figure is more than a polished image.
AI can improve visual structure. Scientific accuracy and submission readiness remain author decisions.
Scientific meaning
AI-assisted pass
Surface unclear relationships and visually ambiguous directionality.
Author check
Confirm every entity and relationship against the source material.
Terminology
AI-assisted pass
Improve hierarchy, label length, contrast, and placement.
Author check
Approve exact terms, notation, abbreviations, units, and capitalization.
Visual system
AI-assisted pass
Align modules and normalize strokes, spacing, and color roles.
Author check
Confirm that color and shape semantics are consistent and accessible.
Submission fit
AI-assisted pass
Improve legibility and composition at the intended figure size.
Author check
Check journal dimensions, file format, resolution, and final export.
A figure changes at every research handoff.
Carry the same source through lab discussion, coauthor review, reviewer response, and final submission checks.
Lab meeting
Capture the first visual problems while the scientific discussion is still fresh.
Separate the assay setup from the analysis stage.
Coauthor review
Resolve terminology and hierarchy without rebuilding the agreed figure structure.
Use the manuscript terms and preserve every pathway edge.
Reviewer response
Translate a written critique into a bounded, inspectable revision request.
Clarify encoder to decoder flow and enlarge the labels.
Submission check
Review the selected version at its intended size before it leaves the author team.
Verify panel labels, line weights, and final reading order.
V1 scope
Built for schematics, not experimental evidence.
The boundary is part of the product. We focus on explanatory figures where revisions can be inspected against a source.
Designed for
01Methodology and workflow diagrams
02Mechanism and pathway illustrations
03AI and ML architecture figures
04Graphical abstracts
05Conceptual research diagrams
Illustrative scientific schematicFig. 01
Not for
Fabricating or removing experimental features
Cloning regions in microscopy, gels, or medical images
Changing plotted values or underlying measurements
Keep every revision reviewable.
AI can help draft a visual change. The researcher remains responsible for the evidence, labels, relationships, and final figure.
01
Keep the source
The original remains in the conversation beside later versions.
02
Verify scientific meaning
Check terminology, symbols, arrows, and relationships before export or publication.
03
Use the right material
Revise explanatory schematics here. Preserve raw experimental data in an appropriate scientific workflow.
Questions before you revise
What the V1 workflow supports, where it stops, and what researchers still need to verify.
AI Figure Editor is a revision workspace for scientific schematics. Upload an existing figure, describe the requested changes, and generate a new version without discarding the original.
V1 is designed for conceptual diagrams, methodology workflows, mechanism illustrations, graphical abstracts, and AI or ML architecture figures.
Yes. Add an annotation guide and describe the change. The source figure remains the primary input while the marked copy guides the requested region.
No. Generative image models can alter labels, symbols, and scientific meaning. Every result is a draft that must be compared with the source and reviewed by a qualified person.
The product does not support generative alteration of underlying experimental data. Use it for explanatory schematics and non-data figure layout, not for fabricating, deleting, cloning, or selectively enhancing evidence.
No. Your source image and generated results remain in the conversation so you can compare versions and continue from a chosen result.
Yes, the workspace can create a new scientific schematic from a text description, but the V1 product is optimized around revising an existing figure.
Turn revision feedback into a new figure version.
Upload your schematic, describe the change, compare the result, and verify every scientific detail before use.
