ModelsLab

The modelslab provider supports text-to-image generation through ModelsLab's v6 image API, using /api/v6/images/text2img. The model name is sent unchanged as model_id, so use an exact ModelsLab ID supported by that endpoint.

Setup

1. Create an API key at ModelsLab

2. Set the environment variable:

   export MODELSLAB_API_KEY=your_api_key_here

Provider Format

modelslab:image:<model_name>

Featured Models

Text to Image:

• modelslab:image:seedream-5-lite-t2i - ByteDance Seedream 5 Lite, documented in ModelsLab's v6 integration guide
• modelslab:image:flux - Flux, high-quality image generation
• modelslab:image:sdxl - Stable Diffusion XL

info

Browse the model catalog and check each model's API version. For example, Nano Banana 2 uses gemini-3.1-t2i in a v7 example; that does not establish support through this v6 provider.

Environment Variables

Variable	Description
MODELSLAB_API_KEY	Your ModelsLab API key

Configuration

Basic Setup

providers:
  - id: modelslab:image:flux
    config:
      width: 1024
      height: 1024

Configuration Options

Parameter	Type	Default	Description
apiKey	string	-	API key (or use MODELSLAB_API_KEY env)
width	number	512	Image width in pixels
height	number	512	Image height in pixels
num_inference_steps	number	30	Number of denoising steps
guidance_scale	number	7.5	How closely to follow the prompt
samples	number	1	Images requested upstream. Only the first is returned, so values above 1 bill extra for no additional output.
seed	number	-	Random seed for reproducibility
negative_prompt	string	-	What to avoid in the image
safety_checker	string	no	Enable safety filter (yes or no)
enhance_prompt	string	no	Auto-enhance the prompt (yes or no)

The provider returns a Markdown image reference, ![prompt](url). To check that shape, use an assertion with type: contains and value: '!['.

Full Example

promptfooconfig.yaml

# yaml-language-server: $schema=https://promptfoo.dev/config-schema.json
providers:
  - id: modelslab:image:flux
    config:
      width: 1024
      height: 1024
      num_inference_steps: 50
      guidance_scale: 7.5
      negative_prompt: 'blurry, low quality'
      seed: 42

prompts:
  - 'Generate an image of: {{subject}}'

tests:
  - vars:
      subject: 'a mountain landscape at sunset'

Async Generation

ModelsLab uses an async generation pattern. When an image request returns status: "processing", the provider automatically polls the fetch endpoint every 3 seconds until the image is ready (up to 3 minutes).

Authentication

ModelsLab uses key-in-body authentication. The API key is sent as the key field in the JSON request body rather than as a Bearer token header.