On October 22, Ideogram released Vector 1.0, its first model that outputs vector files instead of pixels (RuntimeWire). Ideogram's announcement on X passed 500,000 views in three days, and the reason is simple: logos, icons, stickers and print graphics all end up as vectors, and until now almost every AI image tool handed you a flat PNG.

This post covers what Ideogram Vector does, where it falls short, and how to get vector-ready art today from the raster models on EasyImagen with a few prompt rules and a free tracing step.

What Ideogram Vector does

From the launch coverage and the hosted API listings:

  • Two modes. Text-to-vector creates an SVG from a prompt. Image-to-vector converts an existing image into an SVG (RuntimeWire).
  • Where to use it. In the Ideogram app, through its API, and on partner platforms.
  • Price per file. On fal, a text-to-vector SVG costs $0.13 at low quality, $0.16 at medium and $0.20 at high, or $0.32 at high quality with reference images. Image-to-vector conversion starts at $0.10 per image (RuntimeWire).

The output is a real SVG made of paths you can recolor, resize and edit in Illustrator, Figma or Inkscape. In Masonry's tests, none of the files contained an embedded raster image.

What early tests found

Masonry published six first-hand SVG files from the model and inspected what was inside them. A few findings matter if you plan to use it for real work:

  • Text comes back as outlines, not live type. Headlines are drawn as paths, so a typo cannot be fixed by retyping. You reshape it or regenerate.
  • File size varies a lot. A simple logo was 7 paths and under 5 KB. A medium-quality poster was 2,001 paths and over 1 MB.
  • Photo conversions get flattened. Stone texture became flat regions of color and hair turned into ribbon-like shapes.
  • Seeds did not work in their tests, so every rerun gave a different result.

None of this is unusual for a version 1.0 model. It does tell you the best inputs: simple, flat, graphic designs, not photos.

Getting vector-ready art from any image model

You don't need a vector model to end up with an SVG. The older route still works well: generate a clean raster image, then trace it into paths with Inkscape's free Trace Bitmap, Illustrator's Image Trace, or Ideogram's own image-to-vector endpoint.

Tracing only works when the image is made for it. These prompt rules make the biggest difference:

  1. Flat color only. Say "no gradients, no shadows, no texture, no noise". Every gradient becomes dozens of stepped shapes after tracing.
  2. A small palette. Name two to four colors. Fewer colors give fewer, cleaner paths.
  3. Thick, even shapes. Ask for "thick even line weights" and "smooth geometric curves". Thin hairlines break up when traced.
  4. A plain white background with margins. That way the background drops out in one click and nothing touches the edge.
  5. One subject, or a clearly separated grid. Overlapping objects merge into one shape.

Example 1: a flat logo with GPT Image 2

A flat vector logo for a fictional coffee roaster, centered on a plain pure white background. The mark: a simple mountain peak inside a coffee bean shape, built from 3 solid flat colors only (dark espresso brown, warm orange, cream), thick even line weights, smooth geometric curves, no gradients, no shadows, no texture, no noise. Below the mark, the wordmark "NORTHLINE COFFEE" in a clean bold geometric sans-serif, dark espresso brown, evenly spaced. Generous empty margin around everything. Render exactly this text, spelled correctly, and no other words.

Flat vector-style logo of a mountain peak inside a coffee bean in espresso brown, orange and cream, with the wordmark NORTHLINE COFFEE below

This came out with three solid colors, clean edges and the wordmark spelled correctly. Note that, just like Ideogram Vector, tracing turns the wordmark into outlines. If you need editable type, trace only the mark and set the name yourself in a real font. "Northline Coffee" is a made-up brand. Before you use any AI logo commercially, check that it doesn't resemble an existing trademark.

Example 2: a matching icon set

Exactly six flat vector app icons for a fictional travel app, arranged in two rows of three on a plain pure white background, with wide even gaps between them. Top row, left to right: a suitcase, a paper airplane, a map pin. Bottom row, left to right: a camera, a palm tree, a ticket. Every icon uses the same style: a rounded square tile in soft teal with a solid white glyph centered on it, two colors only, consistent thick shapes, rounded corners, no gradients, no shadows, no texture. Six icons total, each subject appears once. No text, no letters.

Six flat teal app icons in two rows: suitcase, paper airplane, map pin, camera, palm tree and ticket

Grids are where image models slip. Our first attempt, using Nano Banana and a shorter prompt ("6 icons in a 3 by 2 grid"), came back with nine tiles and a duplicated row. For the version above we switched to GPT Image 2, spelled the grid out row by row and added "six icons total, each subject appears once". This is the same plan-first idea we covered in our Canva Craft post: describe the layout explicitly and the model follows it.

Which route should you take?

  • You need an editable SVG straight away and will pay per file: Ideogram Vector is the most direct option, especially for simple marks and icons.
  • You want to explore many ideas cheaply first: generate raster drafts in the EasyImagen generator with the rules above, then trace only the one you pick. Ideogram's image-to-vector endpoint or Inkscape can do that last step.
  • Your design has a lot of text: keep the text out of the traced art and set it in a real font, whichever route you use.

For more ready-to-use prompts, browse the prompt library or the AI tools pages, such as the new Day of the Dead generator with flat folk-art styles that also trace well.