What's new

What changed in the editor, newest first. Something you would like to see here? Tell us.

Paste a picture into the chat

Pasting a screenshot or a logo into the chat used to do nothing at all. Now a picture can be pasted, dropped on the message box or added with the paperclip button. Send it on its own and you choose how it becomes SVG: redrawn as an illustration by the image model, or traced in your browser for free. Send it with a message and the assistant sees it, so you can ask for colours to match it or a logo like it. Traced pictures also keep far more of their detail than before.

Read the article

Move and resize shapes in the preview

The preview could show you a shape's code and delete it, but moving or resizing anything meant editing numbers by hand. Now you can drag a shape to move it and drag its handles to resize it, and the change is written into the code the way you would write it yourself: a circle's cx and r, a rectangle's x and width, a traced path's translate. The rest of the file stays exactly as it was, and one Ctrl+Z undoes a whole drag.

See every change

The assistant can look at your drawing

In a traced picture every shape is an unnamed path, so "make the hat red" had to be worked out from colours and coordinates, and the assistant could repaint the wrong brown. It can now look before it edits: your browser draws the picture with the shapes it suspects outlined in colour, and it changes the ones that sit on the part you named. Icons got a lot better too: they come in at a sensible size, an icon asked for on its own becomes the whole drawing, and brand logos are found rather than redrawn by hand.

Read the article

Export as a picture, a favicon or code

Download used to give you the .svg and nothing else, so a picture for a slide, a favicon for a site or the markup in the shape a web page wants meant taking the drawing to another tool. Download now opens a menu: the SVG file as before, a PNG or WebP picture at any size, a favicon with the app icons that go with it, or the drawing as code — a data URI, a CSS background or a React component. Each one is shown before anything is saved or copied, so what you get is what you saw.

See every change

Recolour one part of a traced drawing

A traced drawing keeps every region of one colour in a single shape, so a tree's trunk and its leaves can be one path. Asked to colour the trunk, the assistant could only recolour that whole path, painting the leaves with it — and asked again, it did the same thing again. It can now see the separate parts inside a path, tell a trunk from a leaf by where each sits and how big it is, and give each part its own path and colour, without moving anything else. This release also makes the chat friendlier around the edges: pasted SVG code opens straight away, a dropped connection says what happened, and the assistant tells you when it had to guess.

See every change

Replies stream in as they are written

A reply used to arrive all at once, after a wait with nothing to watch but "Thinking…" — long enough on a big drawing to wonder whether anything was happening. Replies now appear as they are written, the assistant's reasoning shows while it works, and each tool it calls is named as it starts.

See every change

Claude joins the model picker

Three Claude models are now available to Pro subscribers. Claude Sonnet 5 is the one the picker offers by default, at twenty credits; Claude Haiku 4.5 is quicker and cheaper at five, and Claude Opus 5 is the strongest at thirty, both under "Show all". Sonnet and Opus take the same thinking-effort setting the gpt-5 models do, so you can ask for a quick answer on a small change and a careful one on a drawing that needs thinking about.

See every change

Generated pictures keep their detail

Some drawings the assistant writes as SVG itself; a picture of something — an animal, a scene, a logo — it generates as an image and then traces into shapes. On that second route, the instructions we send along with your words were working against you: they asked for the fewest possible colour regions and forbade fine detail, on the theory that simpler images trace more cleanly. They do — into something much plainer than you asked for. A fox curled asleep in ferns came back as a flat orange disc. Those instructions now ask for flat, cel-shaded colour instead of the least colour possible, and say nothing about how many colours to use, so a busy subject stays busy and a plain one stays plain.

See every change

v2.1

The assistant edits shapes, not lines

Rewriting a line means retyping all of it, coordinates and all, so one slip moves a label — and the edit costs whatever that line is long, which on a traced path is thousands of characters. A minified drawing has no lines to address at all. The assistant can now write only what you asked to change instead, addressed by id, class or position: translating two hundred labels is one call to find them and one to rewrite them, and recolouring a traced drawing is a single edit. Line editing has not gone anywhere — it is still how the assistant works on a document that is mid-edit and not yet valid XML, where elements cannot be addressed at all.

See every change

A rebuilt model line-up

The model picker had grown into a list of near-identical options. It now offers one model per price point, grouped into free and pro tiers, with the credit cost of a request shown next to each.

See every change

Chat history moved to the cloud

AI conversations used to live in the browser, which meant they were gone if you switched device or cleared your data. They are now stored with your account, along with the drawing each message produced — so undo still works after a reload, and an unfinished drawing is waiting for you on the Files page.

See every change

AI features now need an account

Anonymous AI use was being abused faster than it could be paid for. AI features now require signing in, and every account gets a free monthly allowance of credits. The editor itself stays free and needs no account at all.

See every change

v2.0

A rebuilt editor with an AI assistant

The editor was rewritten from the ground up: a resizable live preview, schema-aware autocomplete drawn from the W3C SVG specification, and an AI assistant that edits your drawing from a plain-language instruction — or draws one from scratch and vectorizes it.

See every change