WaveSpeedAI

Recraft V4 Styles on WaveSpeedAI: Create a Style ID, Then Generate With It

Step-by-step guide to the Recraft V4 style workflow on WaveSpeedAI: create-style from reference images, then styled text-to-image and text-to-vector, with the full price ladder.

By WaveSpeedAI7 min read

Recraft V4’s style workflow on WaveSpeedAI is a two-call pattern: run recraft-ai/recraft-v4/create-style on 1 to 10 reference images to get a style_id ($0.0055), then pass that style_id to recraft-ai/recraft-v4-style/text-to-image ($0.0385) or recraft-ai/recraft-v4-style/text-to-vector ($0.055) as many times as you like. A Pro track with the same shape exists at recraft-v4-pro/create-style, recraft-v4-style-pro/text-to-image ($0.11), and recraft-v4-style-pro/text-to-vector ($0.132). All six endpoints went live on August 27, 2026.

This guide walks through the flow with the real parameter names from the schemas, explains the one pricing modifier, and lays out the full Recraft V4 price ladder so you can pick a tier.

What Recraft V4 Styles is

Recraft’s pitch for V4 Styles is “style it once, every image matches”: you supply reference images and the model holds to their rendering technique, color, texture, and composition across every generation (Recraft V4 Styles). Recraft says the feature won nearly 92 percent of side-by-side evaluations against seven competitors; that is a vendor figure, not something we have reproduced.

The mechanics come from Recraft’s own API. A style is built from “one to ten images,” the match field is precise by default or flexible for looser aesthetic matching, and a generation request may carry either a saved style_id or inline style references but not both (Recraft API docs). WaveSpeedAI exposes the same model with the same field semantics, so anything you learn from Recraft’s documentation carries over.

If you are new to the model family, Recraft V4 Text-to-Image on WaveSpeedAI and Recraft V4 Text-to-Vector cover the base endpoints.

Step 1: create a style from reference images

Call recraft-ai/recraft-v4/create-style (or recraft-ai/recraft-v4-pro/create-style for the Pro track). The endpoint type is image-to-text: images in, a style_id string out.

ParameterTypeRequiredDefaultAllowed values
imagesarray of stringsyesnone1 to 10 image URLs
base_stylestringnoanyany, vector_illustration
matchstringnopreciseprecise, flexible

Practical notes from the model readme and Recraft’s docs:

  • Use visually consistent references. Ten images in three different styles produce a muddier style than three images in one style.
  • Set base_style to vector_illustration when the references are flat or vector art and you plan to feed the style to a text-to-vector endpoint. Leave it at any for photographic or painterly references.
  • precise keeps the output close to the references; flexible lets the model generalize. Start with precise and relax only if outputs feel too locked to the reference compositions.
  • Price is a flat $0.0055 per request. images, base_style, and match do not add charges, so a 10-image style costs the same as a 1-image style.
{
  "images": [
    "https://example.com/brand-illustration-01.png",
    "https://example.com/brand-illustration-02.png",
    "https://example.com/brand-illustration-03.png"
  ],
  "base_style": "vector_illustration",
  "match": "precise"
}

Submit that with POST https://api.wavespeed.ai/api/v3/recraft-ai/recraft-v4/create-style, then poll GET /api/v3/predictions/<id>/result. The completed result carries the style_id. Store it; you will reuse it across every generation and it is the only thing you need from this step.

Step 2: generate with the style_id

The four styled generation endpoints share one schema.

ParameterTypeRequiredDefaultAllowed values
promptstringyesnonedescribe the image or vector graphic
style_idstringnononea style_id from the matching create-style endpoint
imagesarray of stringsnononeup to 10 reference URLs, used only when style_id is absent
aspect_ratiostringno1:11:1, 4:3, 3:4, 16:9, 9:16
style_matchstringnopreciseprecise, flexible

Either style_id or images must be present; a request with neither has no style to apply. Note the field name difference: create-style uses match, the generation endpoints use style_match.

{
  "prompt": "A delivery courier on an electric cargo bike passing a row of cafes, morning light",
  "style_id": "STYLE_ID_FROM_STEP_1",
  "aspect_ratio": "16:9",
  "style_match": "precise"
}

Send that to recraft-ai/recraft-v4-style/text-to-image for a raster image or recraft-ai/recraft-v4-style/text-to-vector for SVG output. The same body works on the -pro variants provided the style_id came from recraft-v4-pro/create-style.

Style IDs are not interchangeable across tiers

The model readmes are explicit: a standard Recraft V4 style_id works only with recraft-v4-style models, and a Pro style_id works only with recraft-v4-style-pro models. If you intend to generate on both tiers, run both create-style endpoints against the same reference set and keep two IDs. At $0.0055 each that is a negligible cost.

The shortcut: inline references and the $0.0055 modifier

You can skip step 1 entirely by passing images to a generation endpoint with no style_id. The platform creates a temporary style for that request. This is the only pricing modifier in the workflow. The live formula on all four styled endpoints is:

total_price = base_price + (images supplied and count > 0 ? $0.0055 : $0)

So the inline route costs exactly the base price plus one create-style call, every time. Evaluated for each endpoint:

Endpoint (model_uuid)With style_idWith inline images
recraft-ai/recraft-v4-style/text-to-image$0.0385$0.044
recraft-ai/recraft-v4-style/text-to-vector$0.055$0.0605
recraft-ai/recraft-v4-style-pro/text-to-image$0.11$0.1155
recraft-ai/recraft-v4-style-pro/text-to-vector$0.132$0.1375

The break-even is one generation. If you will use a style twice, create it once. Inline references are for one-off exploration, or for pipelines where a fresh reference set arrives with every job.

The full Recraft V4 price ladder

The styled endpoints sit inside a wider Recraft V4 family on WaveSpeedAI. Prices below are the live base prices; none of the non-style endpoints carries a formula.

Endpoint (model_uuid)OutputPrice
recraft-ai/recraft-v4/create-stylestyle_id$0.0055
recraft-ai/recraft-v4-pro/create-stylestyle_id$0.0055
recraft-ai/recraft-v4-style/text-to-imageraster$0.0385
recraft-ai/recraft-v4/text-to-imageraster$0.04
recraft-ai/recraft-v4.1/text-to-imageraster$0.04
recraft-ai/recraft-v4.1/text-to-image-utilityraster$0.04
recraft-ai/recraft-v4-style/text-to-vectorSVG$0.055
recraft-ai/recraft-v4/text-to-vectorSVG$0.08
recraft-ai/recraft-v4.1/text-to-vectorSVG$0.08
recraft-ai/recraft-v4-style-pro/text-to-imageraster$0.11
recraft-ai/recraft-v4-style-pro/text-to-vectorSVG$0.132
recraft-ai/recraft-v4-pro/text-to-imageraster$0.25
recraft-ai/recraft-v4.1-pro/text-to-imageraster$0.25
recraft-ai/recraft-v4.1-pro/text-to-image-utilityraster$0.25
recraft-ai/recraft-v4-pro/text-to-vectorSVG$0.30
recraft-ai/recraft-v4.1-pro/text-to-vectorSVG$0.30

Two things stand out. First, styled generation is not a premium over the plain endpoints: recraft-v4-style/text-to-image at $0.0385 is slightly below recraft-v4/text-to-image at $0.04, and recraft-v4-style/text-to-vector at $0.055 is well below the $0.08 of recraft-v4/text-to-vector. Second, the Pro style tier ($0.11 and $0.132) is less than half the price of the plain Pro endpoints ($0.25 and $0.30). If you already have a reference look, the style endpoints are the cheaper way into each quality tier.

The trade-off is control surface. The plain V4 and V4 Pro endpoints take a size string (default 1024*1024); the V4.1 family takes an image_size preset; the styled endpoints take only the five aspect_ratio values. The V4.1 utility variants are described by Recraft as faster, lighter models for high-volume raster work at the same price as their non-utility siblings.

Choosing a tier

  • Brand illustration systems, icon sets, repeated social templates: create one standard style, generate on recraft-v4-style/text-to-vector for editable SVG or recraft-v4-style/text-to-image for raster. This is the cheapest consistent-style path on the platform.
  • Client-facing hero art where fidelity matters: create a Pro style and generate on the -style-pro endpoints. You get the Pro model at less than half the plain Pro price.
  • One-off style tests: pass images inline and accept the $0.0055 surcharge.
  • No reference look yet: use the plain recraft-v4 or recraft-v4.1 endpoints, which give you size control and need no setup.

For background on why Recraft V4 is worth this attention, see Recraft V4: How a Small AI Startup Topped Image Generation.

Model pages

Share