Image Tools Guide

Generate, edit, clean up and analyse images — all of it on your own machine, with the results landing straight on the canvas.

Last updated — 25 August 2026

Generating an image

Two routes in: describe what you want, or draw it badly and let the model finish it.

From a description

Describe the picture you want in plain language; the model generates it and it lands on the canvas.
Several image models are available, including Google's Nano Banana alongside other options.
Local image models work with no connection at all. The optional Nexus AI cloud module reaches more models and spends Merits.

From a sketch

Draw roughly on the canvas — boxes, arrows, a stick figure — and send the drawing to the model. It interprets the sketch and returns a finished image to the same canvas. This is the fastest route to a diagram you can picture clearly but cannot draw.

Writing an image prompt that works

Name the subject first, then the setting, then the style.
Say what should not be in the picture; models honour an exclusion far better than they honour vagueness.
Change one thing at a time between attempts, so you learn which change did the work.
Keep the prompt that worked. It will be worth more next month than the image was.

Editing an image you already have

Image-to-image editing and direct canvas edits cover different jobs.

Image to image

Feed an existing picture back to the model with an instruction and get a revised version out. This is how you restyle a photograph, change a background, or push a rough diagram towards something cleaner without redrawing it from scratch.

Direct edits on the canvas

Crop, resize, rotate and flip.
Brightness, contrast, saturation and blur.
Greyscale, sepia and invert.
Brush and shape tools to draw over the top, and text overlay with the full font controls.

Cleaning an image up

The tools that make a photograph usable in a document.

Background removal

Background removal lifts the subject out of a photograph so it can sit on your own colour or on another image. It works best on a clearly separated subject. Hair, glass and fine detail at the edges are where it struggles, and a larger, sharper source image always gives it more to work with.

Working non-destructively

Keep the original on the canvas and edit a duplicate, so you can always go back to the untouched version.
Make one change at a time, and judge the result at final size rather than zoomed in.
Undo steps back through the edits: Ctrl + Z, or Command + Z on macOS.
Save the project before a long series of edits, not after.

Analysing an image

The same vision models that read handwriting will answer questions about any picture.

What you can ask for

Explain image: what is in the picture and what it appears to show.
Extract data: text, figures and table contents lifted out as editable text.
Custom prompt: any question of your own about the image.

Reading handwriting

This is what most people come for. Photograph the page evenly lit and square on, crop to the writing, then run the extraction. Handwritten answer sheets can be marked the same way, against your own answer key and rubric, without a single script leaving the machine. The vision settings guide covers the model choice and the temperature to use.

Formats, quality and shortcuts

Small decisions that decide whether the result survives printing.

Choosing a format

PNG when you need transparency, or crisp text and lines.
JPEG for photographs, where file size matters more than the last few percent of quality.
Start from the highest-quality source you have. No model recovers detail that a compressed file already threw away.
Export at the pixel size you will actually use, rather than scaling a small export up afterwards.

Keyboard shortcuts

Ctrl + Z — undo. Ctrl + Y — redo.
Ctrl + S — save. Ctrl + C and Ctrl + V — copy and paste.
B — brush. T — text. C — crop. M — move. S — shape.
On macOS, use Command in place of Ctrl throughout.