infinity-ml 0.1.0

Python client for Infinity Media Server - GPU-accelerated image, video, and audio generation

infinity_ml

Python client for Infinity Media Server - GPU-accelerated image, video, and audio generation.

Designed to mirror the fal_client API for familiarity.

Installation

pip install infinity_ml

Quick Start

Module-level Functions (fal_client style)

from infinity_ml import run, submit, subscribe, status

# Simple blocking call
result = run("black-forest-labs/flux-dev", {"prompt": "A cat astronaut in space"})
print(result["images"][0]["download_url"])

# Submit and poll manually
handle = submit("black-forest-labs/flux-dev", {"prompt": "A cat"})
print(f"Submitted: {handle.request_id}")

# Wait with progress updates
for event in handle.iter_events():
    print(f"Status: {event}")
    
result = handle.get()

# Subscribe with callbacks
def on_update(status):
    print(f"Progress: {status}")

result = subscribe(
    "black-forest-labs/flux-dev",
    {"prompt": "A beautiful sunset"},
    on_queue_update=on_update,
)

Client Class Usage

from infinity_ml import SyncClient, AsyncClient

# Sync client
client = SyncClient()

# Run and wait
result = client.run("black-forest-labs/flux-dev", {"prompt": "A cat"})

# Convenience methods with typed results
image_result = client.image("A futuristic city")
print(image_result.images[0].download_url)

video_result = client.video("Ocean waves at sunset", num_frames=16)
print(video_result.video.download_url)

audio_result = client.audio("Hello from Infinity!")
print(audio_result.audio.download_url)

# Download file
client.download(audio_result.audio.download_url, "output.wav")

Async Usage

import asyncio
from infinity_ml import AsyncClient, run_async, submit_async

async def main():
    # Module-level async functions
    result = await run_async("black-forest-labs/flux-dev", {"prompt": "A cat"})
    
    # Or use AsyncClient
    async with AsyncClient() as client:
        handle = await client.submit("black-forest-labs/flux-dev", {"prompt": "A cat"})
        result = await handle.get()

asyncio.run(main())

API Reference

Module-level Functions

Function	Description
run(app, args)	Submit and wait for result (blocking)
submit(app, args)	Submit job, returns SyncRequestHandle
subscribe(app, args, ...)	Submit with status callbacks
status(app, request_id)	Get job status
result(app, request_id)	Get job result
cancel(app, request_id)	Cancel job (if supported)

Async versions: run_async, submit_async, subscribe_async, status_async, result_async, cancel_async

SyncClient / AsyncClient

client = SyncClient(
    base_url="http://192.222.55.191:8000",  # Default demo server
    default_timeout=300.0,                   # Max wait time (seconds)
)

Methods

Method	Description
run(app, args)	Submit and wait for result
submit(app, args)	Submit job, returns handle
subscribe(app, args, ...)	Submit with callbacks
status(app, request_id)	Get job status
result(app, request_id)	Get job result
get_handle(app, request_id)	Get handle for existing job
health()	Server health check
ready()	Server readiness & GPU status
models()	List available models

Convenience Methods (typed results)

Method	Description
image(prompt, ...)	Generate image, returns ImageResult
video(prompt, ...)	Generate video, returns VideoResult
audio(text, ...)	Generate audio, returns AudioResult
download(url, path)	Download generated file

Request Handles

Returned by submit(), provides methods to track job progress:

handle = submit("black-forest-labs/flux-dev", {"prompt": "A cat"})

handle.request_id           # Job ID string
handle.status()             # Returns Status (Queued, InProgress, Completed, Failed)
handle.get()                # Wait and return result (blocking)
handle.iter_events()        # Iterate status updates until complete
handle.cancel()             # Cancel job (if supported)

Status Types

from infinity_ml import Queued, InProgress, Completed, Failed

status = handle.status()

if isinstance(status, Queued):
    print(f"Queue position: {status.position}")
elif isinstance(status, InProgress):
    print("Processing...")
elif isinstance(status, Completed):
    print("Done!")
elif isinstance(status, Failed):
    print(f"Error: {status.error}")

Available Models

Model ID	Type	Example
black-forest-labs/flux-dev	Image	{"prompt": "A cat"}
wan-ai/wan2.2-t2v	Video (2.2)	{"prompt": "Waves", "guidance_scale": 4.0, "guidance_scale_2": 3.0}
wan-ai/wan2.1-1.3b	Video (2.1)	{"prompt": "Waves", "num_frames": 33}
wan-ai/wan2.1-14b	Video (2.1)	{"prompt": "Waves", "num_frames": 33}
boson-ai/higgs-audio-v2	Audio	{"text": "Hello!"}

See client_model_docs/ for detailed parameter documentation.

Caveats (Demo Version)

⚠️ This is a demo/development version:

• Hardcoded server IP: The default URL (http://192.222.55.191:8000) points to a Lambda Labs development instance. This will be replaced with a proper domain in production.
• No authentication: The demo server has no API key requirement.
• Ephemeral storage: Generated files are stored temporarily and may be cleaned up.

For production use, specify your own server URL:

from infinity_ml import SyncClient

client = SyncClient("https://your-production-server.com")

Or set the INFINITY_SERVER_URL environment variable:

export INFINITY_SERVER_URL=https://your-production-server.com

License

MIT