Krea 2 API overview

Krea 2 is our first foundation image model, trained completely from scratch to give you full control over the look, feel, and creative direction of every image you generate. What sets it apart:

Aesthetic diversity — Krea 2 can generate a wide range of styles, with particular strength in artistic and expressive directions other models tend to flatten.
Style transfer — the most advanced style transfer system available. Extract the style of one or many reference images and apply it to your outputs with precision.
Tunable creativity — creativity is an actual parameter you can adjust. At high values, the model adds aesthetic depth, interpretation, and visual richness to whatever you give it.
Moodboards — pass in an entire moodboard of dozens of images and Krea 2 will understand the overall creative direction, producing outputs that align with it.

Explore the gallery

Hundreds of Krea 2 outputs with their prompts.

Try Krea 2

Krea’s latest image model — styles, LoRAs, and photographic detail.

Choose a variant

The Krea 2 API ships in two variants. Both support the same public request shape — the differences come down to speed, model size, training stage, and the character of the outputs they produce.

Krea 2 Medium

Smaller, faster, more cost-efficient. Extensive post-training makes outputs especially stable and consistent across generations.Strongest on: illustration, anime, painting, and other expressive or artistic styles.

Krea 2 Large

More than 2× the size of Medium, with softer post-training. Outputs are rawer, more textured, and more flexible — at its best, Large produces results Medium can’t match.Strongest on: photorealism, raw looks (motion blur, grain, low dynamic range), expressive and artistic styles.

Not sure which to start with? Use Medium for stable general-purpose output, and Large when you want the richest texture or photorealism.

Pricing

Variant	Text-to-image	With style references	With moodboards
Medium	$0.030	$0.035	$0.040
Large	$0.060	$0.065	$0.070

Combining moodboards with style references does not increase the price per generation — you pay the moodboard rate ($0.040 Medium / $0.070 Large).

Endpoints

Variant	Endpoint
Medium	`POST /generate/image/krea/krea-2/medium`
Large	`POST /generate/image/krea/krea-2/large`

Quickstart

Submit a generation request and poll for the result. All requests require a Bearer token — see API Keys & Billing if you don’t have one yet.

// npm install @krea-ai/sdk
import { Krea } from "@krea-ai/sdk";

const krea = new Krea({ apiKey: process.env.KREA_API_KEY });

const result = await krea.subscribe("image/krea/krea-2/medium", {
  input: {
    prompt: "a cinematic glass cabin beside a frozen lake at sunrise",
    aspect_ratio: "16:9",
    resolution: "1K",
  },
});

// https://gen.krea.ai/images/80ead844-02a6-467d-ba9e-fcd401bcb9a6.png
console.log(result.data?.urls[0]);

curl -X POST https://api.krea.ai/generate/image/krea/krea-2/medium \
  -H "Authorization: Bearer $KREA_API_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "A cinematic product photo of a glass lamp on a marble table",
    "aspect_ratio": "4:5",
    "resolution": "1K",
    "creativity": "medium"
  }'

# Then poll the returned job_id
curl https://api.krea.ai/jobs/<job_id> \
  -H "Authorization: Bearer $KREA_API_TOKEN"

import os
import time
import requests

API_BASE = "https://api.krea.ai"
API_TOKEN = os.environ["KREA_API_TOKEN"]

def wait_for_job(job_id):
    while True:
        response = requests.get(
            f"{API_BASE}/jobs/{job_id}",
            headers={"Authorization": f"Bearer {API_TOKEN}"},
        )
        response.raise_for_status()
        job = response.json()
        if job["status"] == "completed":
            return job
        if job["status"] in ("failed", "cancelled"):
            raise RuntimeError(f"Job {job['status']}")
        time.sleep(3)

response = requests.post(
    f"{API_BASE}/generate/image/krea/krea-2/medium",
    headers={
        "Authorization": f"Bearer {API_TOKEN}",
        "Content-Type": "application/json",
    },
    json={
        "prompt": "A cinematic product photo of a glass lamp on a marble table",
        "aspect_ratio": "4:5",
        "resolution": "1K",
        "creativity": "medium",
    },
)
response.raise_for_status()
job = wait_for_job(response.json()["job_id"])
print(job["result"]["urls"])

package main

import (
	"bytes"
	"encoding/json"
	"fmt"
	"net/http"
	"os"
	"time"
)

const apiBase = "https://api.krea.ai"

func waitForJob(client *http.Client, jobID, token string) (map[string]any, error) {
	for {
		req, _ := http.NewRequest("GET", fmt.Sprintf("%s/jobs/%s", apiBase, jobID), nil)
		req.Header.Set("Authorization", "Bearer "+token)
		resp, err := client.Do(req)
		if err != nil {
			return nil, err
		}
		var job map[string]any
		json.NewDecoder(resp.Body).Decode(&job)
		resp.Body.Close()
		switch job["status"] {
		case "completed":
			return job, nil
		case "failed", "cancelled":
			return nil, fmt.Errorf("job %s", job["status"])
		}
		time.Sleep(3 * time.Second)
	}
}

func main() {
	token := os.Getenv("KREA_API_TOKEN")
	client := &http.Client{}

	payload, _ := json.Marshal(map[string]any{
		"prompt":       "A cinematic product photo of a glass lamp on a marble table",
		"aspect_ratio": "4:5",
		"resolution":   "1K",
		"creativity":   "medium",
	})

	req, _ := http.NewRequest("POST", apiBase+"/generate/image/krea/krea-2/medium", bytes.NewBuffer(payload))
	req.Header.Set("Authorization", "Bearer "+token)
	req.Header.Set("Content-Type", "application/json")

	resp, _ := client.Do(req)
	var created map[string]any
	json.NewDecoder(resp.Body).Decode(&created)
	resp.Body.Close()

	job, err := waitForJob(client, created["job_id"].(string), token)
	if err != nil {
		panic(err)
	}
	fmt.Println(job["result"])
}

Krea 2 Large generations can take longer to complete, so for production workloads prefer the X-Webhook-URL header over polling.

Parameters

krea-2/medium and krea-2/large accept the same public request body.

Field	Type	Notes
`prompt`	string	Required. Text prompt describing the image.
`aspect_ratio`	string	One of `1:1`, `4:3`, `3:2`, `16:9`, `2.35:1`, `4:5`, `2:3`, `9:16`.
`resolution`	string	Currently only `1K` is supported.
`seed`	number	Optional. Use the same seed + prompt to reproduce a generation.
`creativity`	enum	`raw`, `low`, `medium` (default), `high`. See below.
`image_style_references`	array	Optional. See Style transfer.
`moodboards`	array	Optional. See Moodboards. Max 1 moodboard per request.
`styles`	array	Optional trained styles (LoRAs).
`intensity`	integer	Optional. `-100` to `100`, default `0`. See Generative sliders.
`complexity`	integer	Optional. `-100` to `100`, default `0`. See Generative sliders.
`movement`	integer	Optional. `-100` to `100`, default `0`. See Generative sliders.

Creativity

creativity controls how far the model expands on what you wrote. When your prompt is short or vague, the model fills in missing pieces — style, composition, camera angle, color palette — and creativity decides how much liberty it takes.

raw — no expansion

The model renders only what you’ve explicitly described. Best for tightly art-directed prompts where every detail is already specified.

low — close to the prompt

Minimal expansion. The model stays close to the literal prompt but fills in obvious gaps.

medium — balanced (default)

Default behavior. The model adds reasonable interpretation without straying far from the prompt’s intent.

high — expressive interpretation

Strong expansion. The model takes meaningful creative liberty with style, mood, and aesthetics — best for short or open-ended prompts where you want the model to surprise you.

Feature deep-dives

Style transfer

Extract the style of one or many reference images and apply it to your outputs with tunable strength.

Moodboards

Reference a Krea moodboard to set the overall visual direction — palette, texture, mood, and composition.

Generative sliders

Steer intensity, complexity, and movement of every generation without rewriting your prompt.

Open source

Krea 2 is also available as an open-source release if you’d rather self-host or fine-tune the model instead of (or alongside) calling the hosted API. Two checkpoints are published on Hugging Face:

Krea 2 RAW — the undistilled base checkpoint. Diverse and malleable, intended for fine-tuning, post-training, and LoRA training.
Krea 2 Turbo — an 8-step distilled checkpoint built for fast text-to-image inference. LoRAs trained on RAW are designed to apply cleanly on Turbo.

Resource	Link
Checkpoints on Hugging Face	huggingface.co/krea/Krea-2-Raw
Model code on GitHub	github.com/krea-ai/krea-2
Technical report	Krea 2 technical report

Licensing

The open-source release ships under two license tracks:

Community License — free, non-exclusive use for research, personal, and non-commercial projects, plus commercial use under a revenue threshold. Subject to the Acceptable Use Policy.
Commercial License — enterprise terms for commercial use that exceeds the Community License revenue threshold. Contact opensource@krea.ai.

The hosted Krea 2 API documented on this page is governed by the standard API terms — you don’t need to accept the open-source license to call krea-2/medium or krea-2/large.

Getting Started

Krea 2

Core Concepts

Code Examples

Tutorials

Krea 2 API overview

Choose a variant

Krea 2 Medium

Krea 2 Large

Pricing

Endpoints

Quickstart

Parameters

Creativity

Feature deep-dives

Style transfer

Moodboards

Generative sliders

Open source

Licensing

API reference

​Choose a variant

Krea 2 Medium

Krea 2 Large

​Pricing

​Endpoints

​Quickstart

​Parameters

​Creativity

​Feature deep-dives

Style transfer

Moodboards

Generative sliders

​Open source

​Licensing

​API reference

Choose a variant

Pricing

Endpoints

Quickstart

Parameters

Creativity

Feature deep-dives

Open source

Licensing

API reference