What AI has added to Figma, Canva and Photoshop

Every design tool you are likely to open at work has added AI in the last few years, and each launch came with a keynote and a claim that design had changed. Strip out the announcements. Three real things arrived. Selection and masking got dramatically better, so cutting a subject out of a photograph stopped being a craft. Generation arrived inside the files you already had, on a layer beside your own work, which is why people used it whatever the quality of the pictures. And repetitive work became something you describe to the tool instead of scripting, which is the change nobody put in an announcement and the one that gives a small studio back the most hours. This lesson sorts the new features into those three kinds, with real examples from Figma, Canva and Photoshop, so that you can judge the next one yourself.

The first kind is assistive. An assistive feature speeds up a step you were already doing, and it is most of what shipped. In Photoshop, Select Subject finds the person or product in a photograph with one click, where the pen tool used to take an hour on a difficult edge. Figma's Remove background does the same for a picture placed in a layout, and Canva has its own Background Remover beside Magic Grab, which lifts the subject out of a photo so you can move or resize it. None of these makes a decision for you. They take away the tracing, and the judgement of the edge stays with you.

The second kind is generative, which means it makes pixels, layouts or words that were not in the file before. Photoshop's Generative Fill is the best known. You select an area, type a prompt, and it returns a set of variations on a new generative layer that you can hide, swap or delete without touching the photograph underneath. Generative Expand grows the canvas and fills the new strip to match. Figma can make an image from a prompt, edit one or extend it past its edges, and Canva's Magic Media, Magic Expand and Magic Edit do the same jobs for its users. Since late 2025 Photoshop has also offered models from other companies, such as Google's Gemini and Black Forest Labs' FLUX, in the same Generative Fill box as Adobe's own Firefly. These are the features that get the headlines. They are also the least reliable.

The third kind is automation, and it changes a working week while getting the least attention. Resizing forty assets, renaming layers to a convention and pulling copy from a spreadsheet into a template were never interesting work, and they used to eat whole days. Canva's Magic Switch turns one design into other sizes and formats, and it can translate the words into another language on the way. Figma's Rename layers names frames, groups, text and image layers from what they contain and where they sit, leaves alone anything you already named, and costs no credits. Since May 2026 Figma has also had an agent on the canvas, still in beta, that takes a plain instruction for a chore with several steps, such as making every layer name on a page follow one convention. You understand this kind of job completely. That makes the output easy to check.

A working designer sorts each new feature into one of those three boxes before deciding how far to trust it, because each box fails in its own way. Assistive tools fail at the edges, on hair, glass and fur. Generative tools fail by being average. They produce the most likely answer to your request, and the most likely answer can echo the designs the model learnt from. Automation fails quietly, when one file in forty is wrong and nobody opens that one. So the failure tells you the check. You zoom to 200 per cent along a mask, you name the decision a generated layout made for you and ask whether you agree with it, and you open a random sample of a batch instead of only the first file.

Every one of these features now runs on a meter. Figma gives a Full seat on its Professional plan 3,000 AI credits a month, and its price list charges 1 to 5 credits to remove a background and 2 to 16 to make or edit an image, depending on the model. Rename layers and AI search cost nothing. Adobe meters generation in generative credits as well, and counts the partner models in Photoshop as premium. In practice the three boxes sort the bill too, because tidying is often free while making new images is where an allowance goes. One setting deserves a look before any client file goes in. Since 15 August 2024, Figma has been allowed to use files from Starter and Professional teams to train its AI models unless an admin turns content training off, and turning it off leaves every AI feature working.