Kolors Image Generation

Kuaishou Kolors model — ultra-low-cost image generation with Chinese and English prompt support, multiple aspect ratios, and batch generation. Ideal for high-volume generation and testing.

Try it Online

Test in Playground ↗ · View Models ↗

Quick Start

from openai import OpenAI

client = OpenAI(
    base_url="https://open.dieyuyun.com/v1",
    api_key="YOUR_API_KEY"
)

response = client.images.generate(
    model="Kwai-Kolors/Kolors",
    prompt="An orange tabby cat sitting on a windowsill watching the rain, watercolor style, warm tones",
    size="1024x1024",
    n=1
)

print(response.data[0].url)

Endpoint

Item	Value
Method	POST
Path	/v1/images/generations
Base URL	https://open.dieyuyun.com
Protocol	OpenAI Images API

Authentication

Bearer Token: Authorization: Bearer YOUR_API_KEY

Standard Request Fields

Field	Type	Required	Default	Description
model	string	Yes	—	Model ID: Kwai-Kolors/Kolors
prompt	string	Yes	—	Image description (CN/EN supported)
size	string	No	1024x1024	Image dimensions (width×height)
n	integer	No	1	Number of images (range 1-4)

Supported Sizes

Size	Ratio
1024x1024	1:1
768x1024	3:4
1024x768	4:3
720x1280	9:16
1280x720	16:9

Extended Parameters

Field	Type	Required	Default	Description
negative_prompt	string	No	—	Unwanted elements description
seed	integer	No	—	Random seed for reproducibility (0-9999999999)
num_inference_steps	integer	No	20	Inference steps — more = higher quality (1-100)
guidance_scale	number	No	7.5	Prompt adherence strength (0-20)

Request Examples

cURL

curl https://open.dieyuyun.com/v1/images/generations \
  -H "Authorization: Bearer $API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "Kwai-Kolors/Kolors",
    "prompt": "An orange tabby cat sitting on a windowsill watching the rain, watercolor style, warm tones",
    "size": "1024x1024",
    "n": 1
  }'

Python

from openai import OpenAI

client = OpenAI(
    base_url="https://open.dieyuyun.com/v1",
    api_key="YOUR_API_KEY"
)

response = client.images.generate(
    model="Kwai-Kolors/Kolors",
    prompt="An orange tabby cat sitting on a windowsill watching the rain, watercolor style, warm tones",
    size="1024x1024",
    n=1
)

print(response.data[0].url)

Node.js

import OpenAI from 'openai'

const client = new OpenAI({
  baseURL: 'https://open.dieyuyun.com/v1',
  apiKey: 'YOUR_API_KEY',
})

const response = await client.images.generate({
  model: 'Kwai-Kolors/Kolors',
  prompt: 'An orange tabby cat sitting on a windowsill watching the rain, watercolor style, warm tones',
  size: '1024x1024',
  n: 1,
})

console.log(response.data[0].url)

Response Format

Success Response

{
  "created": 1717000000,
  "data": [
    {
      "url": "https://..."
    }
  ],
  "usage": {
    "generated_images": 1,
    "total_tokens": 0
  }
}

Field	Type	Description
created	integer	Unix timestamp
data	array	Generated results array
data[].url	string	Image URL
usage	object	Usage statistics
usage.generated_images	integer	Number of generated images
usage.total_tokens	integer	Total tokens consumed

Error Response

See Error Codes.

Key Features

Ultra-Low Cost

At just ¥0.01 per image, Kolors is the most affordable image generation model on the platform — perfect for high-volume generation and testing.

Batch Generation

Generate up to 4 images per request with the n parameter, ideal for comparing multiple variations.

response = client.images.generate(
    model="Kwai-Kolors/Kolors",
    prompt="Modern minimalist living room design",
    size="1024x1024",
    n=4
)

for img in response.data:
    print(img.url)

Fine-Grained Control

Use guidance_scale and num_inference_steps to precisely tune generation quality:

response = client.images.generate(
    model="Kwai-Kolors/Kolors",
    prompt="Cyberpunk city at night",
    size="1280x720",
    guidance_scale=12.0,
    num_inference_steps=50
)

Best Practices

• Batch generation: Use n to generate multiple candidates and pick the best one.
• Negative prompts: Use negative_prompt to exclude unwanted elements, e.g. "low quality, blurry, deformed".
• Prompt tips: Supports mixed Chinese and English prompts. The more specific the description, the better the result.
• Reproducibility: Save the seed value to reproduce similar results for iterative refinement.
• Step tuning: Default 20 steps is sufficient for most cases. Increase to 50-100 for higher quality.

Rate Limits

See Rate Limits.

Related Docs

• Doubao Seedream 5.0 Lite - High-quality generation with web search
• BFL FLUX - BFL FLUX official model
• Model List & Pricing - Browse available models and pricing