AI Scientific Illustration
Software for Researchers

A contrastive learning diagram with two input modalities. One branch takes a stack of clinical text reports (short echocardiography findings) and passes them through a text encoder. The other branch takes a stack of echocardiogram image frames and passes them through an image encoder. Both encoders project into a shared embedding space. On the right, show two 5×5 similarity matrices side by side: the first labeled "Correlation between embeddings" with a fuzzy, imperfect pattern (current training state), the second labeled "Desired correlation" with a sharp, clean diagonal (training target). Style: flat, editorial, Nature-journal aesthetic. Input cards can be stacked to suggest batches. Composition, palette, and typography are up to the agent — prioritise visual clarity and whitespace.

A two-row comparison diagram contrasting a conventional workflow with a robotics-and-ML workflow for discovering a natural-component plastic substitute. On the far left, a single icon labelled "Natural components" shows a cluster of leaves and biomolecules, representing the shared starting point. On the far right, a photograph-style panel shows the final result: an "All-natural plastic substitute" film with a small scale bar. Between the two ends, two parallel horizontal paths run from left to right. The top path, labelled "Conventional design workflow", is rendered greyed out and shows three sequential steps: laborious experiment, trial-and-error cycles, and experience-based design. The bottom path, labelled "Robotics/ML-integrated workflow", is rendered in a bold accent style and shows an automated pipetting robot feeding into an active learning loop (circular arrow), which in turn feeds an AI/ML prediction network. Arrows flow along each path toward the final product. Style: flat, editorial, Nature-journal aesthetic. Composition and palette are up to the agent.

A six-panel process diagram for fabricating a gold nanocone array, arranged as two rows of three panels with arrows threading through them. Each panel shows an isometric cross-section of the same silicon substrate at a different fabrication stage. Panel 1 "Gold Evaporation" shows a Si/SiO2 slab with a uniform Au film on top. Panel 2 "E-beam Lithography" adds a PMMA resist layer with a regular pattern of circular holes. Panel 3 "Alumina Evaporation" adds an Al2O3 layer over the resist, filling the holes. Panel 4 "Lift-Off" removes the resist, leaving an array of Al2O3 discs on the Au film. Panel 5 "Argon Ion Milling" shows argon ions (Ar+) bombarding the surface from above, eroding the Au around each disc. Panel 6 "Final Structure" shows a clean array of pointed gold nanocones on the silicon substrate. Arrows flow from panel 1 to 6. Style: flat, editorial, Nature-journal aesthetic. Composition and palette are up to the agent.

A vertical four-step molecular mechanism showing how BRCA2 loads RAD51 onto damaged DNA. Step 1 shows a long DNA strand with an RPA-coated single-stranded region on the left and a nucleosome-bound double-stranded region on the right; free RAD51 molecules float above. Step 2 shows a BRCA2 complex (with PALB2, BRCA1, BARD1 partners) approaching the DNA in two ways: a curved arrow labelled "direct binding" pointing to the RPA-coated ssDNA, and a second arrow labelled "diffusion-assisted binding" landing on the dsDNA arm. Step 3 shows the BRCA2-RAD51 complex sliding along the dsDNA toward the junction, with a nucleosome acting as a "sliding barrier" that stops it. Step 4 shows "RAD51 nucleation and stabilization": clustered RAD51 monomers at the ssDNA-dsDNA junction. A final panel at the bottom shows a fully extended "RAD51 filament" along the ssDNA. Vertical arrows connect the steps. Style: flat, editorial, Nature-journal aesthetic. Composition and palette are up to the agent.

A left-to-right model diagram for multi-view animal segmentation. On the far left, show a cylindrical arena labelled "Single animal" containing a small mouse, surrounded by four camera icons aimed inward to suggest "Video capture". An arrow leads to the middle block labelled "Well-trained VisTR", drawn as a stylised transformer architecture: a stack of convolutional feature layers on the left followed by a grid of transformer encoder-decoder blocks on the right, ending in a small output head. Another arrow leads to the right block labelled "Multiview segmentation", showing a 2x2 grid of four viewpoint crops of the mouse with a teal segmentation mask overlaid on each. A final arrow exits to the right to suggest downstream 3D reconstruction. Style: flat, editorial, Nature-journal aesthetic. Keep the arena, network, and segmentation outputs visually distinct. Composition and palette are up to the agent.

A two-panel mechanism illustration showing how gut-derived tryptophan metabolism influences scleral collagen synthesis in high myopia. On the left, a cartoon human torso highlights the intestine in pink, with an arrow pointing up to a dashed callout containing a pair of eye cross-sections. The left eye shows a normal-length eyeball with upward arrows next to "SP1" and "col1a1" (indicating active SP1 binding to the COL1A1 promoter and strong collagen production, drawn as a DNA helix with a protein icon). The right eye shows an elongated, myopic eyeball with downward arrows next to the same labels, indicating reduced SP1 binding and decreased collagen. Below the eyes, a horizontal band depicts the gut epithelial lining: on the left, tryptophan metabolism produces abundant 3-IAA; on the right, dysbiosis leads to reduced 3-IAA. Vertical arrows connect the gut panel up to the eye panel on each side, showing how plasma 3-IAA drives the sclera phenotype. Style: flat, editorial, Nature-journal aesthetic. Composition and palette are up to the agent.

A two-row workflow diagram for a microbiome GWAS study. The top row shows the pipeline left to right in three stages. Stage one "Cohorts" lists four human cohorts (Dutch DMP, LLD, 500FG, 300OB, and a Tanzanian cohort) with small icons and sample sizes. Stage two "Metagenomic sequencing and human SNP genotyping" shows a stool-sample icon feeding into a sequencing reads panel, paired with a DNA-helix icon feeding into a SNP genotyping grid. Stage three "Microbial SV detection" shows a small participant-by-genes matrix with vSV and dSV columns. The bottom row shows three result panels side by side: a boxplot of vSV abundance stratified by human SNP genotype (AA/AC/CC), a bar chart of dSV presence rate by genotype, and a Manhattan plot titled "GWAS on microbial SVs" with chromosomes on the x-axis and log10 p-values on the y-axis. Style: flat, editorial, Nature-journal aesthetic. Composition, palette, and typography are up to the agent.

A four-stage left-to-right diagram of a liquid-crystal-network coating as it is chemically converted. Each stage shows two things: a macroscopic view of a thin yellow coating on a glass slide (with a small zoom marker), and a molecular inset below it showing the mesogen arrangement. Stage I "Initial smectic coating" shows a pristine polymer film with ordered stacks of rod-shaped mesogens (acrylates, azobenzenes, benzoic acid units, and porogen fillers) arranged in neat smectic layers. Stage II "After porogen removal" — connected by an arrow labelled "Solvent extraction" — shows the same film minus the porogen rods, leaving gaps. Stage III "Polymer salt formation" — connected by "Alkaline bath" — shows the benzoic acid groups converted to charged benzoate salts. Stage IV "Refilled with liquid" — connected by "Refilling" — shows the film immersed in a solvent bath with small solvent molecules filling the voids. A small legend at the bottom identifies each mesogen type. Style: flat, editorial, Nature-journal aesthetic. Composition and palette are up to the agent.