OpenAI API image generation endpoint#

API Reference#

OpenVINO Model Server includes now the images/generations endpoint using OpenAI API. Please see the OpenAI API Reference for more information on the API. The endpoint is exposed via a path:

http://server_name:port/v3/images/generations

Example request#

curl http://localhost:8000/v3/images/generations \
  -H "Content-Type: application/json" \
  -d '{
    "model": "black-forest-labs/FLUX.1-schnell",
    "prompt": "three cats",
    "num_inference_steps": 10,
    "size": "512x512"
  }'| jq -r '.data[0].b64_json' | base64 --decode > output.png

Example response#

{
  "data": [
    {
      "b64_json": "..."
    }
  ]
}

Request#

Param

OpenVINO Model Server

OpenAI /images/generations API

Type

Description

model

✅

✅

string (required)

Name of the model to use. Name assigned to a MediaPipe graph configured to schedule generation using desired embedding model. Note: This can also be omitted to fall back to URI based routing. Read more on routing topic TODO

prompt

✅

✅

string (required)

A text description of the desired image(s). TODO: Length restrictions? Too short/too large?

size

✅

✅

string or null (default: auto)

The size of the generated images. Must be in WxH format, example: 1024x768. Default model W/H will be used when using auto.

n

❌

✅

integer or null (default: 1)

A number of images to generate. If you want to generate multiple images for the same combination of generation parameters and text prompts, you can use this parameter for better performance as internally computations will be performed with batch for Unet / Transformer models and text embeddings tensors will also be computed only once. Not supported for now.

background

❌

✅

string or null (default: auto)

Allows to set transparency for the background of the generated image(s). Not supported for now.

style

❌

✅

string or null (default: vivid)

The style of the generated images. Recognized OpenAI settings, but not supported: vivid, natural.

moderation

❌

✅

string (default: auto)

Control the content-moderation level for images generated by endpoint. Either low or auto. Not supported for now.

output_compression

❌

✅

integer or null (default: 100)

The compression level (0-100%) for the generated images. Not supported for now.

quality

❌

✅

string or null (default: auto)

Quality of the image that will be generated. Recognized OpenAI qualities, but currently not supported: auto, high, medium, low, hd, standard

response_format

⚠️

✅

string or null (default: b64_json)

The format of the images in output. Recognized options: b64_json or url. Only b64_json is supported for now (default).

user

❌

✅

string (optional)

A unique identifier representing your end-user that allows to detect abuse. Not supported for now.

Parameters supported via extra_body field#

Param

OpenVINO Model Server

OpenAI /images/generations API

Type

Description

prompt_2

✅

❌

string (optional)

Prompt 2 for models which have at least two text encoders (SDXL/SD3/FLUX).

prompt_3

✅

❌

string (optional)

Prompt 3 for models which have at least three text encoders (SD3).

negative_prompt

✅

❌

string (optional)

Negative prompt for models which support negative prompt (SD/SDXL/SD3).

negative_prompt_2

✅

❌

string (optional)

Negative prompt 2 for models which support negative prompt (SDXL/SD3).

negative_prompt_3

✅

❌

string (optional)

Negative prompt 3 for models which support negative prompt (SD3).

num_images_per_prompt

❌

❌

integer (default: 1)

The same as base parameter n.

guidance_scale

✅

❌

float (optional)

Guidance scale parameter which controls how model sticks to text embeddings generated by text encoders within a pipeline. Higher value of guidance scale moves image generation towards text embeddings, but resulting image will be less natural and more augmented.

strength

✅

❌

float (optional) min: 0.0, max: 1.0

Only for image editing endpoints. Indicates extent to transform the reference image. Must be between 0 and 1. image is used as a starting point and more noise is added the higher the strength. The number of denoising steps depends on the amount of noise initially added. When strength is 1, added noise is maximum and the denoising process runs for the full number of iterations specified in num_inference_steps/n. A value of 1 essentially ignores image parameter.

rng_seed

✅

❌

integer (optional)

Seed for random generator.

max_sequence_length

✅

❌

integer (optional)

This parameters limits max sequence length for T5 encoder for SD3 and FLUX models. T5 tokenizer output is padded with pad tokens to ‘max_sequence_length’ within a pipeline. So, for better performance, you can specify this parameter to lower value to speed-up T5 encoder inference as well as inference of transformer denoising model. For optimal performance it can be set to a number of tokens for prompt_3 / negative_prompt_3 for SD3 or prompt_2 for FLUX.

Response#

Param

OpenVINO Model Server

OpenAI /images/generations API

Type

Description

data

✅

✅

array

A list of generated images.

data.b64_json

✅

✅

string

The base64-encoded JSON of the generated image.

data.url

❌

✅

string

The URL of the generated image if response_format is set to url. Unsupported for now.

data.revised_prompt

❌

✅

string

The revised prompt that was used to generate the image. Unsupported for now.

usage

❌

✅

dictionary

Info about assessed tokens. Unsupported for now.

created

❌

✅

string

The Unix timestamp (in seconds) of when the image was created. Unsupported for now.

Currently unsupported endpoints:#

  • images/edits

  • images/variations

Error handling#

Endpoint can raise an error related to incorrect request in the following conditions:

  • Incorrect format of any of the fields based on the schema

  • Tokenized prompt exceeds the maximum length of the model context.

  • Model does not support requested width and height

  • Administrator defined min/max parameter value requirements are not met

References#

End to end demo with image generation endpoint

Code snippets