Skip to content

text_to_image.md

Latest commit

 

History

History
129 lines (103 loc) · 7.82 KB

text_to_image.md

File metadata and controls

129 lines (103 loc) · 7.82 KB

Image Inference API

Generate images from text prompts or transform existing ones using Runware's API. This powerful feature allows you to create high-quality visuals, bringing creative ideas to life or enhancing existing images with new styles or subjects.

Introduction

Image inference enables you to:

  1. Text-to-Image: Generate images from descriptive text prompts.
  2. Image-to-Image: Transform existing images, controlling the strength of the transformation.
  3. Inpainting: Replace parts of an image with new content.
  4. Outpainting: Extend the boundaries of an image with new content.

Advanced features include:

  • ControlNet: Precise control over image generation using additional input conditions.
  • LoRA: Adapt models to specific styles or tasks.

Our API is optimized for speed and efficiency, powered by our Sonic Inference Engine.

Request

Requests are sent as an array of objects, each representing a specific task. Here's the basic structure for an image inference task:

[
  {
    "taskType": "imageInference",
    "taskUUID": "string",
    "outputType": "string",
    "outputFormat": "string",
    "positivePrompt": "string",
    "negativePrompt": "string",
    "height": int,
    "width": int,
    "model": "string",
    "steps": int,
    "CFGScale": float,
    "numberResults": int
  }
]

Parameters

Parameter Type Required Description
taskType string Yes Must be set to "imageInference" for this task.
taskUUID string (UUID v4) Yes Unique identifier for the task, used to match async responses.
outputType string No Specifies the output format: "base64Data", "dataURI", or "URL" (default).
outputFormat string No Specifies the image format: "JPG" (default), "PNG", or "WEBP".
positivePrompt string Yes Text instruction guiding the image generation (4-2000 characters).
negativePrompt string No Text instruction to avoid certain elements in the image (4-2000 characters).
height integer Yes Height of the generated image (512-2048, must be divisible by 64).
width integer Yes Width of the generated image (512-2048, must be divisible by 64).
model string Yes AIR identifier of the model to use.
steps integer No Number of inference steps (1-100, default 20).
CFGScale float No Guidance scale for prompt adherence (0-30, default 7).
numberResults integer No Number of images to generate (default 1).

Additional parameters:

Parameter Type Required Description
uploadEndpoint string No URL to upload the generated image using HTTP PUT.
checkNSFW boolean No Enable NSFW content check (adds 0.1s to inference time).
includeCost boolean No Include the cost of the operation in the response.
seedImage string No* Image to use as a starting point (required for Image-to-Image, In/Outpainting).
maskImage string No* Mask image for Inpainting/Outpainting (required for these operations).
strength float No Influence of the seed image (0-1, default 0.8).
scheduler string No Specify a different scheduler (default is model's own scheduler).
seed integer No Seed for reproducible results (1-9223372036854776000).
clipSkip integer No Number of CLIP layers to skip (0-2, default 0).
usePromptWeighting boolean No Enable advanced prompt weighting (adds 0.2s to inference time).

ControlNet

To use ControlNet, include a controlNet array in your request with objects containing:

Parameter Type Required Description
model string Yes AIR identifier of the ControlNet model.
guideImage string Yes Preprocessed guide image (UUID, data URI, base64, or URL).
weight float No Weight of this ControlNet model (0-1).
startStep integer No Step to start applying ControlNet.
endStep integer No Step to stop applying ControlNet.
controlMode string No "prompt", "controlnet", or "balanced".

LoRA

To use LoRA, include a lora array in your request with objects containing:

Parameter Type Required Description
model string Yes AIR identifier of the LoRA model.
weight float No Weight of this LoRA model (default 1).

Response

The API returns results in the following format:

{
  "data": [
    {
      "taskType": "imageInference",
      "taskUUID": "a770f077-f413-47de-9dac-be0b26a35da6",
      "imageUUID": "77da2d99-a6d3-44d9-b8c0-ae9fb06b6200",
      "imageURL": "https://im.runware.ai/image/ws/0.5/ii/a770f077-f413-47de-9dac-be0b26a35da6.jpg",
      "cost": 0.0013
    }
  ]
}

Response Parameters

Parameter Type Description
taskType string Type of the task ("imageInference").
taskUUID string (UUID v4) Unique identifier matching the original request.
imageUUID string (UUID v4) Unique identifier of the generated image.
imageURL string URL to download the image (if outputType is "URL").
imageBase64Data string Base64-encoded image data (if outputType is "base64Data").
imageDataURI string Data URI of the image (if outputType is "dataURI").
NSFWContent boolean Indicates if the image was flagged as NSFW (if checkNSFW was true).
cost float Cost of the operation in USD (if includeCost was true).

Note: The API may return multiple images per message, as they are generated in parallel.