Skip to main content
POST
/
v1
/
images
/
generations
Image Generation
curl --request POST \
  --url https://api.gravitex.ai/v1/images/generations \
  --header 'Authorization: <authorization>' \
  --header 'Content-Type: application/json' \
  --data '
{
  "model": "<string>",
  "prompt": "<string>",
  "response_format": "<string>",
  "contents": [
    {}
  ]
}
'

Documentation Index

Fetch the complete documentation index at: https://docs.gravitex.ai/llms.txt

Use this file to discover all available pages before exploring further.

​
Introduction

The image generation API supports text-to-image, image-to-image, image editing, and more. Through a unified API interface, you can call multiple mainstream image generation models including Gemini, Doubao Seedream, GPT Image, and Tongyi Qianwen.

​
Authentication

​
Authorization
string
required
Bearer Token, e.g. Bearer sk-xxxxxxxxxx

​
Request Parameters

​
model
string
required
Model identifier, supported models include:
  • Gemini series: gemini-2.5-flash-image (Nano Banana), gemini-3-pro-image-preview (Nano Banana Pro), etc.
  • Doubao Seedream series: doubao-seedream-3-0-t2i-250415, doubao-seedream-4-0-250828, doubao-seedream-4-5-251128, doubao-seededit-3-0-i2i-250628, etc.
  • GPT Image series: gpt-image-2, etc.
  • Tongyi Qianwen series: qwen-image-plus, qwen-image-edit-plus, etc.
​
prompt
string
Text prompt for text-to-image generation
​
response_format
string
default:"url"
Response format: b64_json or urlNote: Different models have different support for response_format:
  • Gemini series: Only supports b64_json format, always returns base64-encoded image data regardless of the value passed
  • Doubao Seedream series: Usually returns URL links, response_format parameter may not take effect
  • GPT Image series: Only supports b64_json format, forces base64-encoded image data
  • Tongyi Qianwen series: Supports both b64_json and url, returns the corresponding format based on the parameter value (b64_json will download from URL and convert to base64)
​
contents
array
Multi-turn content for image-to-image or contextual conversation

​
Basic Examples

curl -X POST "https://api.gravitex.ai/v1/images/generations" \
-H "Authorization: Bearer sk-xxxxxxxxxx" \
-H "Content-Type: application/json" \
-d '{
"model": "gemini-2.5-flash-image",
"prompt": "A cute orange kitten sitting in a garden, sunny day, high quality photography",
"size": "16:9",
"quality": "high",
"n": 1,
"temperature":1.1,
"top_p":0.95,
"response_format": "b64_json",
"image_size": "2K",
"mime_type": "image/png",
"response_modalities": "image"
}'

​
Model-Specific Parameters

Different models support different parameters. Below are detailed parameter descriptions for each model:
​
image
string
doubao-seedream-3-0-t2i-250415 does not support this parameter for input images. Supports URL or Base64 encoding. Among them, doubao-seedream-4.5 and doubao-seedream-4.0 support single-image or multi-image input (see the Multi-Image Fusion example), while doubao-seededit-3.0-i2i only supports single-image input.
​
size
string
Image size, supported sizes depend on model version:
  • doubao-seedream-3.0: 1024x1024, 1152x864, 864x1152, 1280x720, 720x1280, 1248x832, 832x1248, 1512x648
  • doubao-seedream-4.0/4.5: 2048x2048, 2304x1728, 1728x2304, 2560x1440, 1440x2560, 2496x1664, 1664x2496, 3024x1296 (2K) or 4096x4096, 4704x3520, 3520x4704, 5504x3040, 3040x5504, 4992x3328, 3328x4992, 6240x2656 (4K)
​
watermark
boolean
default:"false"
Whether to add watermark
​
seed
integer
Random seed for controlling the randomness of generation results. Same seed produces similar results. Range: 0 to 2147483647
​
guidance_scale
number
Guidance scale, controls how closely the generated image matches the prompt. Higher values are stricter, lower values are more free. Recommended range: 1.0-10.0, default: 2.5. Only supported by doubao-seedream-3.0-t2i-250415 and doubao-seededit-3.0-i2i-250628
​
sequential_image_generation
string
Sequential image generation toggle, only supported by doubao-seedream-4.0 and doubao-seedream-4.5:
  • "auto": Enable sequential image generation
  • "disabled": Disable sequential image generation (default)
​
sequential_image_generation_options
object
Sequential image generation configuration options, only effective when sequential_image_generation is "auto":
  • max_images (integer): Maximum number of images, range 1-4, default 4
​
optimize_prompt_options
object
Only doubao-seedream-4.5 (currently only supports standard mode) and doubao-seedream-4.0 support this parameter.
  • mode (string): Optimization mode
    • "standard": Standard mode, higher quality but longer time (default, supported by both 4.0 and 4.5)
    • "fast": Fast mode, shorter time but average quality (only 4.0)

​
Response Format

{
  "code": 200,
  "msg": "Success",
  "data": {
    "data": [
      {
        "url": "",
        "b64_json": "iVBORw0KGgoAAAANSUhEUgAABAAAAAQA...",
        "revised_prompt": ""
      }
    ],
    "created": 1757320007
  }
}

​
Supported Models

​
Gemini Series

Model Name: gemini-2.5-flash-image (Nano Banana) Core Capabilities:
  • ✅ Text-to-image (pure text description generates images)
  • ✅ Image-to-image (single image + text generates new image)
  • ✅ Multi-image-to-one (2-5 images fusion generation)
  • ✅ Multi-turn conversational image generation (contextual continuous modification)
Supported Aspect Ratios: 1:1, 3:2, 2:3, 3:4, 4:3, 4:5, 5:4, 9:16, 16:9, 21:9 Image Input: Supports URL and Base64 format, max 7MB Model Name: gemini-3-pro-image-preview (Nano Banana Pro) Core Capabilities:
  • ✅ Text-to-image (pure text description generates images)
  • ✅ Image-to-image (single image + text generates new image)
  • ✅ Multi-image-to-one (2-5 images fusion generation)
  • ✅ Multi-turn conversational image generation (contextual continuous modification)
  • ✅ Higher quality output
Supported Aspect Ratios: 1:1, 3:2, 2:3, 3:4, 4:3, 4:5, 5:4, 9:16, 16:9, 21:9 Image Input: Supports URL and Base64 format, max 7MB

​
Doubao Seedream Series

Model Name: doubao-seedream-3-0-t2i-250415 Core Capabilities:
  • ✅ Text-to-image (pure text description generates images)
  • ✅ Supports guidance scale adjustment
  • ✅ Supports random seed control
  • ❌ Does not support image-to-image
Supported Sizes: 1024×1024, 1152×864, 864×1152, 1280×720, 720×1280, 1248×832, 832×1248, 1512×648 Model Name: doubao-seedream-4-0-250828 Core Capabilities:
  • ✅ Text-to-image (pure text description generates images)
  • ✅ Image-to-image (single image + text generates new image)
  • ✅ Multi-image fusion (2-5 images fusion generation)
  • ✅ Sequential image generation
  • ✅ Supports 2K/4K resolution
  • ✅ Supports multiple image formats
Supported Sizes:
  • 2K: 2048×2048, 2304×1728, 1728×2304, 2560×1440, 1440×2560, 2496×1664, 1664×2496, 3024×1296
  • 4K: 4096×4096, 4704×3520, 3520×4704, 5504×3040, 3040×5504, 4992×3328, 3328×4992, 6240×2656
Image Input: Supports JPEG, PNG, WEBP, BMP, TIFF, GIF formats, max 10MB Limitation: Maximum input plus output 15 images, but not recommended to output too many (e.g., 15 text-to-image or 1 input image output 14 images), as more output images take longer time. Model Name: doubao-seedream-4-5-251128 Core Capabilities:
  • ✅ Text-to-image (pure text description generates images)
  • ✅ Image-to-image (single image + text generates new image)
  • ✅ Multi-image fusion (2-5 images fusion generation)
  • ✅ Sequential image generation
  • ✅ Supports 2K/4K resolution
  • ✅ Supports prompt optimization options
  • ✅ Supports multiple image formats
Supported Sizes: Same as doubao-seedream-4-0 Image Input: Supports JPEG, PNG, WEBP, BMP, TIFF, GIF formats, max 10MB Limitation: Maximum input plus output 15 images, but not recommended to output too many (e.g., 15 text-to-image or 1 input image output 14 images), as more output images take longer time. Model Name: doubao-seededit-3-0-i2i-250628 Core Capabilities:
  • ✅ Image editing (single image + text editing)
  • ✅ Supports guidance scale adjustment
  • ✅ Supports random seed control
  • ✅ Image editing (content modification, style transfer, etc.)
  • ❌ Does not support pure text-to-image
Supported Sizes: Adaptive (automatically adjusts based on input image size) Image Input: Supports JPEG, PNG formats, max 10MB, single image

​
GPT Image Generation Series

Model Name: gpt-image-2 Core Capabilities:
  • ✅ Text-to-image (pure text description generates images)
  • ✅ Image-to-image (up to 10 images + text)
  • ✅ Supports image quality selection
  • ✅ Supports input fidelity adjustment
  • ✅ Multi-image fusion generation
Supported Sizes: 1024×1024, 1024×1536, 1536×1024 Image Quality: low, medium, high Generation Count: Can generate 1-10 images per request Image Input: Supports JPEG, PNG, GIF, WEBP formats, max 10MB, up to 10 images

​
Tongyi Qianwen Series

Model Name: qwen-image-plus Core Capabilities:
  • ✅ Text-to-image (pure text description generates images)
  • ✅ Chinese and English text rendering (excels at generating complex text in images)
  • ✅ Multiple artistic styles
  • ✅ Intelligent prompt extension
  • ❌ Does not support image-to-image
Supported Sizes: 1328×1328, 1664×928, 928×1664, 1472×1140, 1140×1472 Billing Method: Per image (¥0.18/image) Model Name: qwen-image-edit-plus Core Capabilities:
  • ✅ Image editing (input one image, output up to 6 images)
  • ✅ Modify text in images
  • ✅ Add/remove/move objects
  • ✅ Transfer image styles
  • ✅ Enhance image details
Image Input: Supports JPEG, JPG, PNG, BMP, TIFF, WEBP formats, max 10MB Output Description: Actual output count is affected by image content and editing complexity, can generate up to 6 images

​
Best Practices

​
Prompt Optimization Tips

  1. Specify aspect ratio needs: Describe composition direction in the prompt
    • Landscape: Use “horizontal composition”, “widescreen view”
    • Portrait: Use “vertical composition”, “vertical view”
  2. High-quality keywords:
    • “high quality”, “HD”, “professional photography”
    • “8k resolution”, “rich details”
  3. Multi-image fusion techniques:
    • Clearly describe the role of each image
    • Specify fusion method (style transfer, element combination, etc.)

​
Size Selection Tips

  • WeChat Moments: 1328×1328 (1:1) or 1140×1472 (3:4)
  • Weibo Header: 1664×928 (16:9)
  • TikTok Cover: 928×1664 (9:16)
  • Xiaohongshu: 1140×1472 (3:4)

​
FAQ

Different models support different formats:
  • Gemini: PNG, JPEG, JPG, WEBP, max 7MB
  • Doubao Seedream 3.0/4.0: JPEG, PNG, max 10MB
  • Doubao Seedream 4.5: JPEG, PNG, WEBP, BMP, TIFF, GIF, max 10MB
  • GPT Image: JPEG, PNG, GIF, WEBP, max 10MB
  • Tongyi Qianwen: JPEG, JPG, PNG, BMP, TIFF, WEBP, max 10MB
Image URLs are valid for approximately 24 hours. It is recommended to download and save immediately after receiving the response, or upload to your own storage service.
Tongyi Qianwen series generates 1 image per request. For multiple images, please make multiple concurrent requests.

Video Generation

View video generation API documentation

Model List

View all supported model information