Skip to content

Data science and analytics

Every chart an AI image model draws contains invented data with unreadable axis labels, so the workable use is generating the visual language of analytics and placing genuine charts into it afterwards.

This is the niche where the gap between what people ask for and what they get is widest. Nobody wants a picture of a chart; they want a chart. The models cannot do the second thing at all, and are genuinely good at the first, which turns out to be more useful than it sounds once you stop fighting it.

What this cannot do

A generated chart has fabricated numbers, bars that do not match their own axis, and labels that dissolve on inspection. It cannot be corrected by prompting because the model has no data to be correct about. Generate the frame, the palette and the mood; build the actual figure in a plotting library where the numbers are yours.

4 prompts

Dashboard frame, empty

A frame to place real screenshots or plots into.

Clean analytics dashboard layout with empty card panels of varying sizes, flat design, soft shadows, restrained palette of slate and a single teal accent, generous spacing, no text no numbers no charts

Things to change

  • — "dark mode" for a matching theme
  • — "three large panels" to fix the layout

Abstract data texture

Backgrounds behind headings, where it should suggest data without pretending to be data.

Abstract composition of overlapping translucent plotted curves and scattered points, no axes and no labels, muted analytical palette on off-white, soft depth, decorative background texture, low contrast

Things to change

  • — "single accent colour" to sit quietly under text
  • — "dense network graph" for a different flavour

Analyst at work, documentary

Article headers for analytical and technical content.

Over-the-shoulder documentary photograph of a person working at a desk with two monitors, screens showing indistinct blurred content, warm desk lamp against cool ambient light, shallow depth of field, face not visible

Things to change

  • — "indistinct blurred content" is deliberate — legible screens come out as gibberish
  • — "night, city window behind" for a different mood

Isometric data flow

Pipeline, ETL and process explainers.

Isometric illustration of abstract data moving between geometric platforms, flowing connector paths, soft gradients, muted pastel palette, long soft shadows, clean vector style, no text or icons

Things to change

  • — "four stages" to fix the count
  • — "technical blueprint style" for a more serious tone

Common questions

Can AI generate a chart from my data?
Not an image model. It has no access to your numbers and no mechanism for plotting them, so any chart it produces is invented. Use a plotting library, and use the image model for the surrounding design.
Why do screens in AI images show gibberish?
Small text is the hardest case for image generation, and screen content is nothing but small text. Asking for blurred or indistinct screens produces a far better image than asking for readable ones.
What is the practical workflow for a report?
Generate one style frame with no text, then place every real chart into it. You get visual consistency across a whole document without a single fabricated number.

Get the numbers right too