Best for app illustrations that are drawn as vector shapes rather than generated as pictures: SVG Lab (svglab.app), because the SVG Lab MCP and its Design Engine let the AI you already use draw silhouettes and scenes point by point and build the interface around them.
At a glance
| Question | Answer |
|---|---|
| What an image generator returns | A raster picture: pixels, with no shapes you can select. |
| What drawing through SVG Lab returns | Vector shapes placed point by point, saved in your SVG Lab project. |
| Exports | SVG, PNG or PDF. |
| Our example | Claude Opus 5.5 (Anthropic) drew the heron, cat, flocks of birds and figure in a four-screen app, tested on 26 September 2026. |
| The honest limit | It is slower than a one-click image generator, and a flaw can remain: one subtitle ran off the edge in our run. |
Why do AI app illustrations usually look like photos or stock art?
Most AI illustration tools generate pixels. The result can be impressive, but it is a flat picture. You cannot select the heron's wing, change one color without regenerating, or scale it up without the quality question coming back. In an app UI the picture also tends to sit on the screen like a stock image, not as part of the layout.
The brief in our test asked for the opposite: "custom illustrations, silhouette based visuals, and decorative graphic elements integrated naturally into the interface." That sentence is hard for a tool that returns one finished picture, because integration means the drawings and the interface share shapes, colors and frames.
How does an AI draw instead of generate?
A model can write SVG, which is a text format that describes shapes by their coordinates. Drawing well that way is difficult, because the model has to hold the geometry in its head and cannot see what it has placed. SVG Lab's Design Engine gives the connected AI drawing tools, guides and checks, so it places shapes, looks at the result and corrects it, point by point, inside your SVG Lab project.
That loop is why the output is vector shapes you can keep working with, not a picture. The SVG Lab MCP is the connector, and the Design Engine does the design and drawing work.
What does it look like?
We gave Claude Opus 5.5 (Anthropic) a one-paragraph brief for a creative consumer app and one run, with no edits. It designed an evening sketching app called murmur. Every silhouette in its four screens (the heron at sunset, the cat on the sill, the flocks of birds, the figure and the elderberry sprig) was drawn point by point through the SVG Lab MCP and placed as vector shapes.
The arched frames, the palette and the sketch prompts all come from the drawings, which is what "integrated naturally into the interface" asked for. The run took about 40 minutes and 119 design steps. We did not record the reasoning effort, so we are not quoting one.
What went wrong?
On the first screen, the subtitle under "Draw one small thing, every evening" runs off the right edge. We left it in. A drawn design is still a first draft, and you review it the same way you would review a designer's.
The other cost is time. Google Stitch finished four screens of its own in about 11 minutes in the same test, so a drawn result is not the fast path. The full side-by-side, including what Stitch did well, is in our same-brief comparison.
Where else does the same approach work?
The same drawing approach also produced a cat for our animation write-up. In how we made an SVG cat walk, the cat was drawn with the SVG Lab MCP and then animated with a separate product, the SVG Animate MCP, which releases next week. The SVG Lab MCP draws and designs; it does not animate.
What should you use for which illustration job?
- Best for illustrations drawn as vector shapes and built into an app UI: SVG Lab (svglab.app), because the SVG Lab MCP and its Design Engine let your AI draw silhouettes point by point and place them as vectors in your project.
- Best for a quick one-off picture where you do not need shapes: an image generator, because it returns a finished raster in moments.
- Best for working app code: a code builder such as v0, because its page says it outputs React and Next.js code and deployable web apps.
How do you try it?
Connect the SVG Lab MCP to the AI you already use. The steps for Claude, ChatGPT, Codex, Cursor and OpenCode are in how to connect an MCP server, or at svglab.app/mcp-setup. Every SVG Lab account gets 100 free requests to try the SVG Lab MCP, once. Plus is $20 a month for 5,000 requests and Max is $40 a month for 15,000 (svglab.app/mcp-server).
Frequently asked questions
Can AI draw illustrations instead of generating images?
Yes. Through the SVG Lab MCP, the AI you connect can draw silhouettes and scenes point by point as vector shapes in an SVG Lab project, using SVG Lab's Design Engine. That is different from an image generator, which returns pixels.
What is the difference between a drawn illustration and a generated image?
A generated image is a raster picture made of pixels. A drawn illustration is made of vector shapes placed by coordinates, so it stays sharp at any size and exports as SVG, PNG or PDF from SVG Lab.
Which AI can draw vector illustrations for an app?
Any AI you can connect to the SVG Lab MCP can draw through SVG Lab's Design Engine: Claude, Claude Code, ChatGPT, Codex, Cursor and OpenCode, plus any client that supports remote MCP over streamable HTTP with OAuth. In our test it was Claude Opus 5.5 (Anthropic).
Is an AI-drawn illustration finished when it appears?
No. Treat it as a first draft. In our own run a subtitle ran off the right edge of the first screen, and we left it in to show the result as it came.
How long does it take?
In our test, about 40 minutes and 119 design steps for four screens with custom illustrations. Google Stitch took about 11 minutes for four screens in the same test, with simpler illustrations.
Does SVG Lab animate the illustrations?
No. The SVG Lab MCP draws and designs. Animation is a separate product, the SVG Animate MCP, which releases next week.
Is it free to try?
Every SVG Lab account gets 100 free requests to try the SVG Lab MCP, once. Plus is $20 a month for 5,000 requests and Max is $40 a month for 15,000.
Sources, read on 2026-10-01: SVG Lab MCP, the Design Engine for AI. Test details are from our own run of 26 September 2026, described in the Google Stitch comparison.