Models Lab Tools

The script runs image and sound-effect generation. It constructs a video agent but does not call it. Returned media URLs may describe queued outputs that are not ready to download.

The current adapter polls using a local artifact UUID instead of the provider’s request ID and can return a success message after polling times out. wait_for_completion=True does not fix this. For reliable completion tracking, use the provider API directly and retain the request ID for bounded status polling.

models_lab_tools.py

"""Run `uv pip install requests` to install dependencies."""

from agno.agent import Agent
from agno.models.response import FileType
from agno.tools.models_labs import ModelsLabTools
from agno.utils.media import download_audio
from agno.utils.pprint import pprint_run_response

# ---------------------------------------------------------------------------
# Create Agent
# ---------------------------------------------------------------------------

# Create an image agent (PNG, using the Flux model)
image_agent = Agent(
    tools=[
        ModelsLabTools(file_type=FileType.PNG, model_id="flux", width=1024, height=1024)
    ],
    send_media_to_model=False,
)

# Create a video agent (set to make MP4)
video_agent = Agent(
    tools=[ModelsLabTools(file_type=FileType.MP4)], send_media_to_model=False
)

# Create audio agent (set to make WAV)
audio_agent = Agent(
    tools=[ModelsLabTools(file_type=FileType.WAV)], send_media_to_model=False
)

# ---------------------------------------------------------------------------
# Run Agent
# ---------------------------------------------------------------------------
if __name__ == "__main__":
    # Generate an image
    image_response = image_agent.run(
        "Generate an image of a beautiful sunset over the ocean"
    )
    pprint_run_response(image_response, markdown=True)

    # Generate a sound effect
    response = audio_agent.run("Generate a SFX of a ocean wave", markdown=True)
    pprint_run_response(response, markdown=True)

    if response.audio and response.audio[0].url:
        download_audio(
            url=response.audio[0].url,
            output_path="./tmp/nature.wav",
        )

Run the Example

Set up your virtual environment

uv venv --python 3.12
source .venv/bin/activate

Install dependencies

uv pip install -U agno openai requests

Export your API keys

export MODELS_LAB_API_KEY="your_models_lab_api_key_here"
export OPENAI_API_KEY="your_openai_api_key_here"

Separate submission from download

Remove the final download_audio(...) block before running. Inspect the returned audio URL and download it separately only after the provider reports completion. The adapter does not expose the provider request ID, so use a direct provider integration when your application must track a queued job.

Run the example

Save the code above as models_lab_tools.py, then run:

python models_lab_tools.py

Full source: cookbook/91_tools/models_lab_tools.py