Ideogram Vector 1.0 Generates Editable SVGs, But the API Has a Layer Caveat

There is a sentence buried in Ideogram's own API reference for its newest model that does not appear in any of the launch coverage. It sits in the parameter list for the vectorizer endpoint of Ideogram Vector 1.0: a boolean called group, described as "Requests grouping into layers," followed immediately by "Grouping is not yet available." That is a remarkable thing to ship in the documentation of a model whose entire pitch is that its output arrives as separate, individually rotatable layers. Ideogram released Vector 1.0 on October 22, 2026 — the company's first dedicated vector model, generating editable SVG from a text prompt and converting uploaded images into vector art — and the shape of what actually works depends heavily on which of the two endpoints you call.

This is not a reason to dismiss the release. It is the reason to read it carefully. Vector 1.0 is a genuinely new capability from a company that built its reputation on getting legible type out of a diffusion model, and moving that strength from pixels into path data is a harder problem than it sounds. But the gap between the marketing summary — components arrive as layers you can move and rotate — and the endpoint-level reality — one path grouped, the other not yet — is exactly the kind of detail that decides whether a design team can build a pipeline on it this month or has to wait.

What shipped on October 22

Two endpoints, both under the same model identifier, ideogram-vector-1, both accepting multipart form data and authenticating with an Api-Key header:

• Text-to-vector — POST https://api.ideogram.ai/v2/svg/generate/ideogram-vector-1. Takes a prompt plus optional reference images and returns SVG artwork.

• Image-to-vector — POST https://api.ideogram.ai/v2/svg/vectorize/ideogram-vector-1. Takes a JPEG, PNG or WEBP up to 50 MB and returns an SVG conversion.

Prompts run from 1 to 9,500 characters, which is generous enough to carry a full brand brief rather than a caption — worth knowing if you were planning to compress your instructions to fit a shorter field.

Both are asynchronous in the way Ideogram's image endpoints are: the call returns a base64 generation ID, and you poll the shared generations endpoint for the finished file. Both accept a webhook URL so you can skip the polling, and those deliveries are Ed25519-signed and verifiable against a published JWKS endpoint — a detail that matters more for a raster-to-vector job than it sounds, because a failed or half-written SVG is worse than no SVG in an automated asset pipeline.

Both also default to private output. Ideogram's docs state that generated outputs are private by default and that enterprise outputs are always private. For logo and brand work, where the asset itself is the confidential thing, that default is the correct one and it is worth noticing that it is on rather than opt-in.

The layer claim, stated precisely

The launch framing is that a Vector 1.0 SVG arrives as separate layers, so each component can be moved, rotated and recoloured independently in a vector editor. As a description of the file you get from the text-to-SVG endpoint, that matches what Ideogram and the outlets covering the launch describe: editable colours, shapes, gradients and typography, structured so that individual elements are addressable rather than fused into one path soup.

The vectorizer endpoint is where the wording needs care. Its group parameter exists specifically to request layer grouping, and the reference page says grouping is not yet available. That means a vectorized image currently comes back as geometry without the layer organisation that makes a generated SVG pleasant to edit. The practical consequence is concrete: if you upload a flat PNG of an existing logo and expect to receive an Illustrator file with a text layer, a mark layer and a background layer, you will not get that today. You will get paths. Whether those paths are cleanly separated by colour and shape even without named groups is a different question, and nobody outside Ideogram has published an inspection of the output yet.

The two modes on the vectorizer are worth knowing for the same reason. Simple "draws fewer, simpler shapes," while the default "preserves more fine detail." That is a straight trade between node count and fidelity, and node count is the number that decides whether a vector file is production-ready or a cleanup job. A traced photograph at default settings can produce thousands of anchor points; the simple mode is the escape hatch. Neither mode is a substitute for grouping, but choosing the right one is the difference between a file a designer can work with and one they redraw.

Pricing, and why you should quote it before you spend it

Ideogram's API reference does not print prices. It handles this with a dry_run query parameter on both endpoints that validates and prices a request without generating, storing, billing or safety-reviewing it, returning a price quote object instead of a generation. That is a genuinely good design for a per-asset model, and it is the authoritative answer for your own account.

What partner rate cards list, and what the launch coverage repeats, is this ladder — treat it as indicative rather than contractual, since it is not printed on Ideogram's own reference page:

• Text-to-SVG quality tiers — $0.13 low, $0.16 medium, $0.20 high, per generated SVG.

• High quality with reference images — $0.32 per SVG. Low and medium stay at their listed rates when references are attached.

• Image-to-vector — $0.10 per converted image.

Read those as per-file prices, not per-iteration prices. A logo exploration where you generate twelve candidates at medium quality costs $1.92 before you have a direction you like; the same exploration on a raster model is fractions of a cent per image and the cost only arrives later, in the designer's time. That is the real shape of the economics and it is why the dry-run endpoint is the most commercially interesting thing in this release: it lets you meter a batch before committing it.

What Vector 1.0 does not do yet

Three gaps are worth stating plainly, because the launch material does not lead with them.

First, layer grouping on vectorization, as above. The parameter is there and the capability is not.

Second, there is no independent quality assessment. Every claim about clean paths, sharp typography and gradient fidelity in circulation right now traces back to Ideogram's own product page or its launch thread. No third party has published a node-count comparison, a path-accuracy measurement, or a blind test of how editable the output actually is. For a raster model that absence is inconvenient; for a vector model it is the whole question, because the difference between a good SVG and a bad one is invisible in a screenshot and painfully obvious the moment you open it.

Third, there is no published spec on maximum output complexity. The generate endpoint's parameter list does not state a limit on num_images, reference images, prompt length or output resolution. The only limit the docs surface is your account's in-flight generation cap, which appears in 429 responses with a reject_reason of inflight_limit and a max_inflight_requests value. If you are planning a batch job, that is the number to discover before you design around throughput.

The advice this release invalidates

Search for how to get vector output from an AI image model and you will find, over and over, the same 2026 consensus: Ideogram is the best typography model but outputs raster only, so route the vector step to Recraft. That guidance was correct when it was written and is now out of date. Ideogram shipping a native SVG model collapses the two-step workflow that advice described — concept in one tool, vector in another — into a single vendor, at least for the cases the model handles well.

It does not make Recraft irrelevant. Recraft's vector output has years of production use behind it, a style system that holds across a set, and export options beyond SVG. What changed is that "Ideogram cannot do vectors" is no longer a true statement, and any comparison page still asserting it should be read as stale. The honest position today is that there are two credible native vector options where there was one, and the tiebreaker is output inspection — which nobody has published for Vector 1.0 yet.

Getting it into a pipeline

OrcaRouter does not serve Ideogram Vector 1.0 — it is not on our catalogue, and we will not pretend otherwise. What we do serve is the rest of the loop around it, and the loop is where vector generation gets awkward in practice. The prompt that produces a good mark is usually written by a language model, not a human at a keyboard; the review step that catches a misspelled wordmark or an off-brand colour is usually a vision model; the routing logic that decides whether an asset goes to a vector model or a raster one is usually a small classifier. Those are the calls that run hundreds of times per batch, and running them on one API for more than 200 models with provider list prices passed through at 0% markup, automatic failover when a provider degrades, and a routing DSL that can put several models behind one endpoint is what makes the expensive per-file call the only one you have to think about.

If you do want to compare output cheaply, the same applies in reverse: the raster models you would benchmark Vector 1.0 against are largely reachable through one key, so building the side-by-side harness does not require a second contract or a second SDK. Failover matters most at exactly the moment a new model is unproven, and this one is two days old.

What to watch

Three things would settle the open question, and none of them exist today.

The first is the group flag turning on. When Ideogram's reference page stops saying grouping is not yet available, vectorized images become editable in the way the launch copy implies, and the endpoint stops being a geometry generator and starts being a layer generator. That is a documentation change, so it will be easy to miss and easy to check.

The second is an independent inspection. Someone needs to take a generated mark, open it in Illustrator or Inkscape, and report the anchor count, the number of distinct paths, whether the type is live text or outlines, and whether the gradients survived. Until that exists, every claim about quality is the vendor's.

The third is a cost benchmark against the alternative. At $0.13 to $0.20 per SVG, Vector 1.0 is priced like a design asset, not like an image. Whether that is cheap or expensive depends entirely on how many of those files a designer would otherwise have to trace by hand — and that number is the one nobody has published either.

For now the accurate summary is this: Ideogram Vector 1.0 is real, it is available on Ideogram's platform and through its API, it does the thing the announcement says for text-to-SVG, and the image-to-vector half is one documentation line short of the same promise. A design team can start testing today. A design team that builds a production pipeline on it this week is betting on a feature flag.