GenMagic for Python and the command line

One API key and one pay-as-you-go balance for 460+ AI models: images, video with sound, speech, music and text. Veo, Kling, Seedance, Wan, Hailuo, Runway, Nano Banana, GPT Image, FLUX, Seedream and more, each with its price shown before you run it.

pip install genmagic
export GENMAGIC_API_KEY=...   # free key, one minute: https://genmagic.co/get-api-key?utm_source=pypi&utm_medium=readme

No dependencies. Python 3.9 or newer.

Images

from genmagic import GenMagic

gm = GenMagic()
image = gm.image("A red fox in snow, 35mm photo", model="google/gemini-nano-banana-2.1", size="1792x1024")
print(image.url, image.cost_cents)
image.save("fox.png")

Video

video starts the job and waits for the clip. Use start_video and get_video to poll yourself.

clip = gm.video("Waves at sunset, slow drone shot", model="alibaba/wan-3.0", aspect_ratio="9:16", duration=5, sound=True)
clip.save("waves.mp4")

Speech, music, text

gm.speech("Welcome to the show").save("welcome.mp3")
gm.music("Calm lo-fi beat, 80 bpm").save("track.mp3")
print(gm.chat("Write a hook for a video about foxes", model="anthropic/claude-haiku-5.5").text)

Tools for AI agents

genmagic.tools holds typed functions with docstrings, which CrewAI, Agno, Pydantic AI, smolagents, Google ADK, Strands and the OpenAI Agents SDK accept as tools:

from genmagic.tools import generate_image, generate_video, text_to_speech, list_models

agent = Agent(tools=[generate_image, generate_video, text_to_speech, list_models])

Models and prices

video_models = gm.models("video")  # needs no key
balance = gm.usage()

A refusal raises GenMagicError with status, code and GenMagic's own message (for example how to top up when the balance is too low).

Command line

genmagic image "A red fox in snow" --out fox.png
genmagic video "Waves at sunset" --aspect 9:16 --duration 5 --out waves.mp4
genmagic speech "Welcome to the show" --out welcome.mp3
genmagic models video
genmagic balance

Also from GenMagic

The same key works in n8n, Zapier, Make, ComfyUI and through MCP in Claude, ChatGPT and Cursor. The REST API is OpenAI-compatible: reference.