Recraft V4 Styles

Recraft V4 Styles is a dedicated model line for style-consistent generation, released in August 2026, and the state of the art in style matching. You define a style once, from one reference image or up to ten, and every generation after that holds to it — rendering technique, color, texture, and composition, not just the general impression. There is no training or fine-tuning step: you upload the reference images, receive a style_id, and generate with it immediately. The references can come from anywhere — other generators, branding materials, or existing assets.

V4 Styles models always require a style. Every request must supply either style_id or style reference images. A request with neither is rejected.

Models

Model	`model` value	Output	Price per image
Recraft V4 Styles	`recraftv4_styles`	Raster, 1K	$0.035
Recraft V4 Styles Vector	`recraftv4_styles_vector`	Vector, SVG	$0.05
Recraft V4 Styles Pro	`recraftv4_styles_pro`	Raster, 2K	$0.10
Recraft V4 Styles Pro Vector	`recraftv4_styles_pro_vector`	Vector, SVG	$0.12

Like the rest of the V4 line, V4 Styles is available in standard (1K) versions for everyday work and fast iteration, and Pro (2K) versions for print-ready assets and large-scale use. Both share the same style behavior.

How to call it

There are two ways to work with the model. Use the two-step flow when a style is reused across many requests — the typical case for brand and campaign work. Use the one-step flow to go straight from references to an image.

Option 1: create a style, then generate

Create the style

Upload your reference images to /v1/styles. The response returns a style_id you can reuse indefinitely.

Generate with that style_id

Pass style_id plus a V4 Styles model to /v1/images/generations. Repeat for every asset in the set.

Step 1: create the style

response = client.post(
    path='/styles',
    cast_to=object,
    options={'headers': {'Content-Type': 'multipart/form-data'}},
    body={'model': 'recraftv4_styles'},
    files={'file1': open('reference-1.png', 'rb')},
)
style_id = response['id']

response = client.post(
    path='/styles',
    cast_to=object,
    body={
        'model': 'recraftv4_styles',
        'image_urls': [
            'https://example.com/reference-1.png',
            'https://example.com/reference-2.png',
        ],
    },
)
style_id = response['id']

curl -X POST https://external.api.recraft.ai/v1/styles \
  -H "Authorization: Bearer $RECRAFT_API_TOKEN" \
  -F "model=recraftv4_styles" \
  -F "file1=@reference-1.png"

curl -X POST https://external.api.recraft.ai/v1/styles \
  -H "Authorization: Bearer $RECRAFT_API_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "recraftv4_styles",
    "image_urls": [
      "https://example.com/reference-1.png",
      "https://example.com/reference-2.png"
    ]
  }'

Response:

{
    "id": "229b2a75-05e4-4580-85f9-b47ee521a00d",
    "style": "any",
    "creation_time": "2026-08-20T00:00:00Z",
    "is_private": true,
    "credits": 5
}

Step 2: generate against the style

response = client.images.generate(
    prompt='product hero shot, bottle on a table',
    model='recraftv4_styles',
    extra_body={'style_id': style_id},
)
print(response.data[0].url)

curl -X POST https://external.api.recraft.ai/v1/images/generations \
  -H "Authorization: Bearer $RECRAFT_API_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "product hero shot, bottle on a table",
    "model": "recraftv4_styles",
    "style_id": "229b2a75-05e4-4580-85f9-b47ee521a00d"
  }'

Option 2: attach style references to the generation

Skip style creation and attach the references to the generation call itself. The server creates a private style from them, applies it, and returns the resolved style_id so later requests can reuse it.

response = client.post(
    path='/images/generations',
    cast_to=object,
    options={'headers': {'Content-Type': 'multipart/form-data'}},
    body={'prompt': 'product hero shot, bottle on a table', 'model': 'recraftv4_styles'},
    files={'style_references': open('reference-1.png', 'rb')},
)
print(response['data'][0]['url'])
print(response['style_id'])

response = client.post(
    path='/images/generations',
    cast_to=object,
    body={
        'prompt': 'product hero shot, bottle on a table',
        'model': 'recraftv4_styles',
        'style_reference_urls': ['https://example.com/reference-1.png'],
    },
)
print(response['data'][0]['url'])
print(response['style_id'])

curl -X POST https://external.api.recraft.ai/v1/images/generations \
  -H "Authorization: Bearer $RECRAFT_API_TOKEN" \
  -F "prompt=product hero shot, bottle on a table" \
  -F "model=recraftv4_styles" \
  -F "style_references=@reference-1.png"

curl -X POST https://external.api.recraft.ai/v1/images/generations \
  -H "Authorization: Bearer $RECRAFT_API_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "product hero shot, bottle on a table",
    "model": "recraftv4_styles",
    "style_reference_urls": ["https://example.com/reference-1.png"]
  }'

style_id and style references are mutually exclusive. Sending both in one request is rejected.

Reference images

Constraint	Value
Formats	PNG, JPG, WEBP
Count	1 to 10
Total size	64 MB across all images, and under 10 MB per file
Transport	`files` / `style_references` for multipart, `image_urls` / `style_reference_urls` for JSON

A single reference is enough. Adding more changes the result in a predictable way: similar references sharpen the match, varied references widen the range. Unlike most models, which only work reliably with references created inside that same model, V4 Styles accepts references from anywhere: other generators, branding materials, or existing assets. There is no requirement that a reference was produced by the model you are generating with. The Create style endpoint also accepts image_weights: per-image weights, one per uploaded image, to bias the style toward specific references.

Style match

Every style stores a match value, set with the match parameter at style creation, that controls how closely generated images follow the style:

precise (the default) — follows the style meticulously, holding every detail: rendering technique, color, composition, and lighting.
flexible — matches the general vibe, giving the model slightly more liberty — for when the mood matters more than the exact blueprint.

Passing style_match in a generation request overrides the stored value for that style. See Style match for details.

Example

Two generations with the same style references — one precise, one flexible:

precise = client.post(
    path='/images/generations',
    cast_to=object,
    options={'headers': {'Content-Type': 'multipart/form-data'}},
    body={
        'prompt': 'product hero shot, bottle on a table',
        'model': 'recraftv4_styles',
        'style_match': 'precise',
    },
    files={'style_references': open('reference-1.png', 'rb')},
)

flexible = client.post(
    path='/images/generations',
    cast_to=object,
    options={'headers': {'Content-Type': 'multipart/form-data'}},
    body={
        'prompt': 'product hero shot, bottle on a table',
        'model': 'recraftv4_styles',
        'style_match': 'flexible',
    },
    files={'style_references': open('reference-1.png', 'rb')},
)

print(precise['data'][0]['url'])
print(flexible['data'][0]['url'])

precise = client.post(
    path='/images/generations',
    cast_to=object,
    body={
        'prompt': 'product hero shot, bottle on a table',
        'model': 'recraftv4_styles',
        'style_reference_urls': ['https://example.com/reference-1.png'],
        'style_match': 'precise',
    },
)

flexible = client.post(
    path='/images/generations',
    cast_to=object,
    body={
        'prompt': 'product hero shot, bottle on a table',
        'model': 'recraftv4_styles',
        'style_reference_urls': ['https://example.com/reference-1.png'],
        'style_match': 'flexible',
    },
)

print(precise['data'][0]['url'])
print(flexible['data'][0]['url'])

# precise: follows the style meticulously
curl -X POST https://external.api.recraft.ai/v1/images/generations \
  -H "Authorization: Bearer $RECRAFT_API_TOKEN" \
  -F "prompt=product hero shot, bottle on a table" \
  -F "model=recraftv4_styles" \
  -F "style_match=precise" \
  -F "style_references=@reference-1.png"

# flexible: matches the general vibe
curl -X POST https://external.api.recraft.ai/v1/images/generations \
  -H "Authorization: Bearer $RECRAFT_API_TOKEN" \
  -F "prompt=product hero shot, bottle on a table" \
  -F "model=recraftv4_styles" \
  -F "style_match=flexible" \
  -F "style_references=@reference-1.png"

# precise: follows the style meticulously
curl -X POST https://external.api.recraft.ai/v1/images/generations \
  -H "Authorization: Bearer $RECRAFT_API_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "product hero shot, bottle on a table",
    "model": "recraftv4_styles",
    "style_match": "precise",
    "style_reference_urls": ["https://example.com/reference-1.png"]
  }'

# flexible: matches the general vibe
curl -X POST https://external.api.recraft.ai/v1/images/generations \
  -H "Authorization: Bearer $RECRAFT_API_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "product hero shot, bottle on a table",
    "model": "recraftv4_styles",
    "style_match": "flexible",
    "style_reference_urls": ["https://example.com/reference-1.png"]
  }'

Pro or Vector generation

Style-consistent generation is not limited to the standard raster model. The line includes Pro (2K raster) and Vector (SVG) variants, and a style applies the same way on each of them: use recraftv4_styles_pro for print-ready 2K raster assets, and recraftv4_styles_vector or recraftv4_styles_pro_vector for editable SVG output.

Example

Generating an SVG with the Pro Vector model and style references:

response = client.post(
    path='/images/generations',
    cast_to=object,
    options={'headers': {'Content-Type': 'multipart/form-data'}},
    body={
        'prompt': 'coffee shop logo, cup with steam',
        'model': 'recraftv4_styles_pro_vector',
    },
    files={'style_references': open('reference-1.png', 'rb')},
)
print(response['data'][0]['url'])
print(response['style_id'])

response = client.post(
    path='/images/generations',
    cast_to=object,
    body={
        'prompt': 'coffee shop logo, cup with steam',
        'model': 'recraftv4_styles_pro_vector',
        'style_reference_urls': ['https://example.com/reference-1.png'],
    },
)
print(response['data'][0]['url'])
print(response['style_id'])

curl -X POST https://external.api.recraft.ai/v1/images/generations \
  -H "Authorization: Bearer $RECRAFT_API_TOKEN" \
  -F "prompt=coffee shop logo, cup with steam" \
  -F "model=recraftv4_styles_pro_vector" \
  -F "style_references=@reference-1.png"

curl -X POST https://external.api.recraft.ai/v1/images/generations \
  -H "Authorization: Bearer $RECRAFT_API_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "coffee shop logo, cup with steam",
    "model": "recraftv4_styles_pro_vector",
    "style_reference_urls": ["https://example.com/reference-1.png"]
  }'

Supported operations

Generate image — a style is always required.
Create style — recraftv4_styles is the default model when no model is specified.

Pricing

Generation and style creation are billed separately. Among the models that support styles, the V4 Styles line has the most attractive pricing for style-consistent generation.

Service	Cost	API units	Basis
Raster generation, Recraft V4 Styles	$0.035	35	Per image
Raster generation, Recraft V4 Styles Pro	$0.10	100	Per image
Vector generation, Recraft V4 Styles Vector	$0.05	50	Per image
Vector generation, Recraft V4 Styles Pro Vector	$0.12	120	Per image
Style creation	$0.005	5	Per request

API units are purchased in advance: $1.00 buys 1,000 units. Composite billing on attached references. When you attach style references to a generation instead of passing a style_id, the style creation is charged once on top of the per-image generation cost. The credits field in the response is the sum of both. Worked example. A style created once, then a 10-image set generation:

	Standard	Pro
Style creation	$0.005	$0.005
10 raster images	$0.35	$1.00
Total	$0.355	$1.005

Reusing that style_id on later sets costs generation only, with no further style-creation charge.

Styles on other V4 models

Styles are not limited to the V4 Styles models. Every V4 and V4.1 model accepts a style_id or attached style references on /v1/images/generations. Generation stays at that model’s own price: applying a style does not change it. Style creation is billed separately at $0.005 per request either way. The difference is that V4 Styles is purpose-built for style consistency and always requires a style, while the other V4 models treat a style as an optional input. See Pricing for generation prices of the other models.

Errors and rejections

Condition	Result
V4 Styles model with no `style_id` and no references	Rejected
`style_id` and style references in the same request	Rejected
More than 10 reference images, or over 64 MB total	Rejected
Reference in a format other than PNG, JPG, WEBP	Rejected
Style references with no `model` specified	Accepted, defaults to `recraftv4_styles`

Requests with style references and no model default to recraftv4_styles. To generate with a different model, set model explicitly.

Getting started

Authentication and your first generation request.

Endpoints

Full parameter reference for generate image and create style.

Styles

How styles work across models: custom styles and style match.

Pricing

Complete API unit charges across every model and operation.

​Models

​How to call it

​Option 1: create a style, then generate

​Option 2: attach style references to the generation

​Reference images

​Style match

​Example

​Pro or Vector generation

​Example

​Supported operations

​Pricing

​Styles on other V4 models

​Errors and rejections

​Related

Getting started

Endpoints

Styles

Pricing

Models

How to call it

Option 1: create a style, then generate

Option 2: attach style references to the generation

Reference images

Style match

Example

Pro or Vector generation

Example

Supported operations

Pricing

Styles on other V4 models

Errors and rejections

Related