Every AI image tool before this week delivered pixels. Ideogram changed that on October 22 when it shipped Vector 1.0, a model that generates real, editable SVG files from a text prompt or an uploaded raster image. This is not a PNG with a vector veneer. The output is actual SVG: paths, gradients, and individually selectable shapes that open in Figma, Illustrator, or your code editor without conversion. Adobe Firefly has been trying to ship text-to-vector generation since 2023. A smaller, specialized AI team got there first.
What the SVG Actually Contains
The distinction matters because “AI-generated vector” has been marketing language before. Independent testing on Vector 1.0 confirmed the output is genuine: zero embedded raster data, zero base64 payloads, and zero live <text> elements in any tested file. Text renders as outlined paths — editable geometry, not invisible font dependencies that break when you move the file.
File complexity scales with subject matter. A simple logo comes back around 4,700 bytes with seven paths. A detailed poster runs to 1.1 MB with roughly 2,000 paths. Each shape is an independent SVG element: rotate it, recolor it, delete it, or hand it off to an LLM to modify as code. SVG is text, so AI coding agents can edit it directly without exporting anything.
Two Endpoints, Two Different Jobs
The Ideogram API has two distinct endpoints. Text-to-vector uses POST /v2/svg/generate/ideogram-vector-1 and takes a prompt up to 9,500 characters plus optional image references. Image-to-vector uses POST /v2/svg/vectorize/ideogram-vector-1 and accepts JPEG, PNG, or WEBP files up to 50 MB, returning SVG with each shape as a selectable element.
Authentication uses the standard Ideogram Api-Key header. The default rate limit is ten concurrent in-flight requests. Both endpoints accept a dry_run flag that returns a cost estimate without billing — the right pattern for batch pipelines where you need to project costs before committing a run.
Pricing and the Recommended Workflow
Pricing is per file, not per token. Text-to-vector costs $0.13 at low quality, $0.16 at medium (the default), and $0.20 at high. Adding image references at high quality pushes it to $0.32. Image-to-vector is a flat $0.10.
The most cost-effective approach separates iteration from vectorization. Ideogram 4.5 — the raster model — supports multi-turn editing at $0.03–0.06 per image. Generate and refine your concept there, then run the winning design through the image-to-vector endpoint once. Total per asset: roughly $0.13–0.16. You pay for exploration at raster prices, then convert only when the design is locked.
Where It Falls Short
Three gaps are worth knowing before you build a pipeline around this. First, there is no edit endpoint. Once a vector file is generated, you cannot ask the API to revise it — you generate again from scratch. That rules out iterative client work where designs evolve through feedback rounds.
Second, the vectorizer does not yet support layer grouping. Files from the image-to-vector endpoint arrive as flat path collections rather than separated, named components. That is manageable for simple marks but becomes a liability for complex logos that need layer-by-layer editing in Illustrator.
Third, seed-based reproducibility is unproven. Independent testing found the seed parameter rejected on multiple attempts, meaning there is currently no reliable way to regenerate the same output. Plan for variation.
The Right Use Cases
Vector 1.0 is well-suited to logos, icon families, packaging artwork, print files, and embroidery assets — anything where the downstream deliverable is a scalable, editable file. It is a poor fit for photographic content, iterative refinement work, and scene composition tasks where Ideogram 4.5 is the better tool.
This is a first-generation model in a new category. The gaps are real, but the trajectory is steep — Ideogram solved a problem Adobe could not ship for three years. A missing layer grouping option is a very different kind of obstacle. Expect these rough edges to close fast. The early independent testing confirms the fundamentals are sound.

