Can ChatGPT make SVG files? What it can do, what it cannot, and the way around it
ChatGPT can write SVG code and it can generate pictures, but it cannot turn the picture it generated into a vector file. What each request really does, and how to get an SVG that matches your image.
Short answer: yes and no. ChatGPT can write SVG code from a description, and the result is a real vector file you can open in Figma or Illustrator. It can also generate an image of a logo. What it cannot do is take that generated image and give you the SVG of it. The two abilities live in two different models, and the gap between them is where most people get stuck.
What 'make an SVG' can mean
- Write SVG code from words: 'a blue circle with a white lightning bolt'. This works, and for simple geometric icons it works well.
- Generate a picture of a logo: this is the image model (DALL·E or the newer image generator inside ChatGPT). The output is a PNG, always.
- Convert that PNG into SVG: this is what people actually want, and it is the part that does not work.
Why the SVG never matches the picture
When you upload the logo it just made and ask for 'the SVG version', the language model does not export anything. It looks at a downscaled copy of the bitmap and types new SVG code from a rough impression of it. Proportions drift, a curve becomes a straight segment, the letter shapes turn into a generic font, the colours are approximations. Ask again and you get a different wrong version, because every answer is a fresh guess, not a measurement.
This is not a temporary limitation of one model version. Writing the coordinates of a Bézier curve by looking at a picture is something no designer can do either; they draw with a tool and adjust until it matches. A text model gets one pass and no way to check.
When asking for SVG code is the right move
If your mark is made of a few geometric primitives, describe it in words and ask for SVG code with a viewBox and named colours. You get tidy, editable output, often cleaner than any conversion. Add 'use currentColor for the main shape' if it is an icon you will recolour from CSS. The moment the design has custom letterforms, a drawn mascot, a gradient or anything with a specific curve, stop: you will spend an hour on revisions and still not have the picture you approved.
How to get the SVG of the image ChatGPT made
- Download the largest PNG the generator offers. Ask for a plain white or transparent background, flat colours, no shadows or 3D effects.
- Convert it with a vectoriser that reconstructs shapes rather than tracing pixels. svgize runs in your browser, measures the slight waviness every AI render carries, and fairs it out while keeping the design.
- Read the numbers next to the preview: a clean logo is a handful of paths and a few colours. Zoom to ×4 and check the corners.
- Download the SVG; PDF export is there for printers that insist on it.
Convert your ChatGPT image to SVG
A note on 'vector' as a prompt word
Prompting the image generator for 'a vector logo, flat, SVG style' changes the look of the picture, not its format. You still get a PNG. The word helps anyway: flat colours and clean edges are exactly what converts best, so keep using it, just do not expect a file with paths at the end.
Questions
Can ChatGPT create vector images?
It can write SVG code from a text description, which is a vector. The pictures it generates with its image model are PNG bitmaps, not vectors, whatever the prompt says.
Can ChatGPT convert a PNG to SVG?
No. Asked for the SVG of an uploaded image, it writes new code from a rough impression of the picture. The result is valid SVG but does not match the image. Use a vectoriser for that step.
Can ChatGPT vectorize a logo?
Not faithfully. For a simple geometric mark, describe it in words and ask for SVG code. For a logo that exists as an image, convert the image with a vectoriser such as svgize.
Is there a ChatGPT SVG generator?
There is no export button. The closest thing is asking for SVG code from a description. For an image you already have, the generator is a vectoriser, not the chat model.
Does the same apply to Midjourney and DALL·E?
Yes. Every image generator available today outputs pixels. Midjourney has no SVG output at all; DALL·E and the ChatGPT image model output PNG. Converting is a separate step for all of them.