Masonry Logo
AI & Technology

Ideogram Vector on Masonry: Real SVG Files From Prompts and Photos

Ideogram Vector is a new image route on Masonry that returns editable SVG files instead of pixels. Inspect three first-hand SVGs, the headline that came back letter-perfect, measured path counts, and how an attached image switches the route to conversion.

Gaurav BisenGaurav Bisen
7 min read

Ideogram Vector is live on Masonry as ideogram-vector, and it is the second route on the platform that returns real SVG files. Send a prompt with no image and it runs text-to-vector, generating logos, posters, and illustrations as editable vector markup. Attach one image and the same key switches to image-to-vector, converting that image into vector shapes; the prompt is not used for a conversion. Both endpoints hand back an image/svg+xml file with the name vector.svg.

The useful part of a launch post is not the format claim. It is three files that were generated, downloaded, opened, and measured, plus the one behaviour that will actually change how you queue work: whether an image is attached decides which endpoint runs.

Evidence boundary: this article rests on six SVG generations across the two endpoints on October 22, 2026, three of which are shown below. Each is a single run of a single input. There were no repeated trials, no held-out prompt set, no standardized queue timing, no reliability rate, and no leaderboard position. File facts came from the fal responses and from opening the downloaded SVGs; path counts came from counting <path> elements. Masonry bills each SVG in credits and shows the cost before each generation; the price comes from fal list rates checked October 23, 2026, set by quality for text-to-vector and flat for image-to-vector. Nothing here establishes how the route behaves on your prompts, your photos, or your acceptance bar.

Quick answer

  • New route: ideogram-vector. A prompt with no image runs text-to-vector; one attached image switches the call to image-to-vector.
  • Returned format: image/svg+xml on all six files: zero embedded raster images, zero base64 payloads, zero <text> elements. Headlines arrive as outlined paths.
  • Measured files: a 4,739-byte logo with 7 paths at 2048×2048; a 1.1 MB poster with 2,001 paths at 1536×2048; a 681 KB photo conversion with 581 paths at 1024×1024.
  • Seed: the published schema lists one; fal rejected it on all four attempts on October 22, 2026, and responses returned null. Reruns vary.
  • Not established here: multi-image batches, quality-tier gaps, latency, reliability, or any Masonry credit price.

The two endpoints and what they accept

ItemChecked October 22, 2026Production implication
Route keyideogram-vectorPin the exact key in automation; the raster Ideogram routes are different models.
Text-to-vector inputRuns when no image is attached. Prompt (required), quality low/medium/high, 17 aspect ratios from 1:4 to 4:1 plus autoNo reference images: an attached image switches the call to image-to-vector.
Image-to-vector inputRuns when one image is attached. JPEG/PNG/WebP up to 50 MB; prompt, quality, and aspect are not sentThe SVG keeps the input image's frame.
Outputimages[] with URL, content_type, file name vector.svg, file sizeDownload promptly; fal file URLs expire.
Reproducibility fieldNone (seed rejected, response seed null)Generate candidates and pick; there is no number to hold.

The image rows deserve emphasis because they change what an attachment means. On a route such as GPT Image 2, an attached image is a reference that steers the prompt. Here it is the input to a conversion: the call switches to image-to-vector, the prompt is not sent, and the SVG redraws the attached image. To generate from a prompt, send it without an image.

The seed row deserves emphasis too, because the schema says otherwise. The fal API schema documents a seed integer from 0 to 2147483647. On October 22, 2026, four text-to-vector submissions carrying a seed — 7 three times, 0 once — each came back input_value_error within a second, while the identical prompt without the field succeeded. Rerun scripts should not assume a seed exists.

Three shipped SVGs

A logo in under 5 KB

Text-to-vector logo, quality low, 1:1. Render of the delivered SVG: 2048×2048, 4,739 bytes, 7 paths.

The brief asked for a minimal geometric coffee bean from two interlocking arcs, in terracotta and cream, with no text. The file is seven paths: background, two bean halves, shading shapes. Nothing to delete, nothing to merge. That sparseness is the specific thing worth checking in a logo run, because a 400-path mark with overlapping fills is a cleanup job, not a deliverable.

A headline that survived contact with vector

Text-to-vector poster, quality medium, 3:4. Render of the delivered SVG: 1536×2048, 1,145,633 bytes, 2,001 paths. Headline verified letter by letter.

The brief put the headline in quotes and named the style, border, and background. HARVEST MARKET came back with all fourteen letters correct in this single run. The aspect behaved too: a 3:4 request returned exactly 1536×2048.

Read the fine print of that success. The letters are outlines, not live type: there is no <text> element in the file, so a typo cannot be fixed by retyping; it can only be reshaped or regenerated. Spell-check the brief before submitting, then proof the render character by character the same way our text-rendering test proofs raster headlines. Vector does not remove the proof step; it removes the retracing step.

A photograph redrawn as 581 paths

The input photo submitted to the image-to-vector route: 1024×1024 PNG.
Image-to-vector conversion. Render of the delivered SVG: 1024×1024, 681,300 bytes, 581 paths.

The conversion keeps what matters for downstream artwork: the face reads as the same person, the blouse print survives as lavender sprigs, the cafe chairs and awning sit where they should. Detail collapses the way posterization always collapses it: stone texture becomes flat regions, hair becomes ribbon shapes. Judge it as a starting point for merch or editorial illustration, the way you would judge a junior designer's first pass: composition and identity preserved, finish still yours to add.

Choosing between the vector routes

For finished pixel images with typography: Ideogram V4, covered in our text-rendering test and our prompting guide.

For anything that must ship as paths: Ideogram Vector. The decision rule is the deliverable format, not taste. If the asset ends in a layout tool as editable shapes (logo, icon set, signage, packaging direction), generate it here. If it ends as a flat export (hero image, ad creative, photoreal scene), generate it on a raster route.

This article ran no auto-tracer, so it makes no path-quality claim against tracing a raster image. The practical test is to open both files in a vector tool and count what you would delete; the logo above has 7 paths. For prompt construction that transfers across either route, the discipline in the prompting guide still applies: name the subject, constrain the palette, quote the copy.

Not established here

  • Batches of more than one SVG per request. Every run here asked for 1.
  • Quality-tier gaps. Low and medium each ran on different briefs, which cannot separate tier from prompt.
  • Latency, queue time, throughput, retry rate, or failure rate. (For the record, fal reported 31.8 seconds of inference time on the poster run: one observation, not a benchmark.)
  • Identity or product fidelity across repeated conversions.
  • Any Masonry credit price, margin, or per-generation cost to you.

Bottom line

One key, two endpoints chosen by whether an image is attached, and one file format that matters: real SVG with no embedded pixels, headlines that survive as outlines, and no seed to lean on. The three files here show the route working across the jobs vector work actually is: a featherweight mark, a typographic poster, a photo redrawn as shapes. They do not rank the route against anything.

Start with the live route, quote the headline, cap the palette, measure the delivered file, and compare cost per accepted asset rather than cost per generation. The full spec sheet with prompt examples lives on Ideogram Vector's model page. Browse the rest of the lineup on Masonry's model hub.

Share:
FAQ

Questions from this guide

Concise answers to the questions readers ask after this guide

What does Ideogram Vector return on Masonry?

Editable SVG files, not pixels. One route key, ideogram-vector, covers both jobs: a prompt with no image runs text-to-vector and generates SVG from the prompt, while an attached image switches the call to image-to-vector, which converts that image into vector shapes and does not use the prompt. Every file measured for this article arrived as image/svg+xml with real paths and no embedded raster data.

Is the HARVEST MARKET headline really spelled right?

Yes, all fourteen letters, in the single run behind this article. The poster brief asked for that exact headline in a bold hand-lettered serif and the returned 1536×2048 SVG draws it as vector outlines. One run is not an accuracy rate, so keep the character-by-character proof before anything goes to print.

Can a seed pin the result for reruns?

No. The endpoint schema published on fal.ai lists a seed field, but fal rejected seeded text-to-vector calls with an input validation error across four runs on October 22, 2026, and responses returned a null seed. Generate candidates and pick rather than trying to reproduce a number.

How heavy are the SVG files?

They track artwork complexity. The measured files run from 4,739 bytes and 7 paths for a simple logo mark, through 681 KB and 581 paths for a photo conversion, to 1.1 MB and 2,001 paths for a detailed illustrated poster. Simple logo and icon work stays featherweight.

When should a team pick Vector over Ideogram V4?

When the deliverable must be editable paths. V4 remains the pick for finished pixel images with strong typography. Vector is the pick for logos, icon sets, signage, and anything that scales, because its lettering arrives as shapes instead of pixels that need retracing. Prove the typographic direction on either route first.

What does image-to-vector do with a photograph?

It redraws the photo as flat vector regions. The conversion in this article took a 1024×1024 street photograph and returned a 1024×1024 SVG with 581 paths, keeping the face, clothing print, and cafe background readable. It is a starting point for merch or editorial illustration, not a finished trace to ship unreviewed.