Quickstart

Get up and running with the multimedia generation umbrella in minutes.

---

Install

uv add lexigram-multimedia

lexigram-multimedia pulls in lexigram, lexigram-contracts, and all seven sibling extension packages (lexigram-multimedia-tts, -music, -video, -image, -upscale, -interpolate, -beat) plus lexigram-tasks and lexigram-storage.

Backend features are opt-in per subsystem — install the extra for the backend you need:

uv add "lexigram-multimedia-tts[elevenlabs,openai]"
uv add "lexigram-multimedia-music[ace-step-server]"
uv add "lexigram-multimedia-video[wan22-server,cogvideox-server]"

With zero extras, the umbrella uses the default in-process/local-http backends.

---

Basic Usage

import asyncio

from lexigram import Application
from lexigram.contracts.multimedia import TTSProvider, TTSRequest
from lexigram.multimedia import MultimediaConfig, MultimediaModule


async def main() -> None:
    config = MultimediaConfig(cache_results=False)
    app = Application(name="my-media-app")
    app.add_module(MultimediaModule.configure(config=config))

    await app.start()

    tts = await app.container.resolve(TTSProvider)
    result = await tts.generate(TTSRequest(text="Hello from Lexigram"))
    if result.is_ok():
        asset = result.unwrap()  # MediaAsset — bytes or a URI
        if asset.has_uri:
            print(f"Generated audio: {asset.uri}")

    await app.stop()


asyncio.run(main())

Note: MultimediaModule.configure() with no arguments is zero-config — it constructs a default MultimediaConfig() internally. Use no-arg for the fastest path.

---

What Just Happened

1. MultimediaModule.configure() returns a DynamicModule owning one MultimediaProvider and exporting all seven subsystem protocols (TTSProvider, MusicProvider, VideoProvider, ImageProvider, UpscaleProvider, InterpolationProvider, BeatAnalysisProvider).
2. On app.start(), MultimediaProvider.register() binds each sub-provider (AudioTTSProvider, AudioMusicProvider, VideoGenerationProvider, ImageGenerationProvider, UpscaleGenerationProvider, InterpolationGenerationProvider, BeatAnalysisGenerationProvider) and delegates to their register() — that is what puts TTSProvider into the container.
3. On boot(), the provider cleanly resolves optional BlobStoreProtocol, CacheBackendProtocol, and EventBusProtocol so nothing breaks when those subsystems are absent, then wires the task queue.
4. container.resolve(TTSProvider) returns the concrete TTS backend. Calling generate() returns Result[MediaAsset, MultimediaError] — never raises for expected failures.

---

Next Steps

• Guide — the accessor model, sync vs queued jobs, timeline composition
• How-Tos — TTS, music, video, upscale, interpolate, beat, compose recipes
• Configuration — the multimedia: config tree and env vars
• Architecture — provider wiring and extension points