Configuration

Configuration options for lexigram-multimedia — the umbrella config plus per-subsystem trees.

---

Overview

Loaded from the multimedia: section of application.yaml. Environment variable prefix: LEX_MULTIMEDIA__.

MultimediaConfig (extends lexigram.config.BaseConfig) declares config_section = "multimedia", so framework config loading maps the YAML/evn-tree to this dataclass. The seven subsystem configs are nested dataclasses — each sibling package owns its own config model and its own defaults; the umbrella simply composes them.

multimedia:
  storage_path_prefix: "multimedia/"
  cache_results: false
  tts:
    backend: "local-http"
  music:
    backend: "local-http"
  video:
    backend: "local-http"
  image:
    backend: "local-http"
  upscale:
    backend: "real-esrgan"
  interpolate:
    backend: "rife"
  beat:
    backend: "librosa"

---

Basic Example

from lexigram.multimedia import MultimediaConfig, MultimediaModule

module = MultimediaModule.configure(
    config=MultimediaConfig(
        storage_path_prefix="media/",
        cache_results=True,
    )
)
app.add_module(module)

Calling MultimediaModule.configure() with no arguments builds the same default configs.

---

Options

Top-level MultimediaConfig

Field	Type	Default	Description
tts	TTSConfig	TTSConfig()	TTS subsystem config (lexigram-multimedia-tts)
music	MusicConfig	MusicConfig()	Music generation config (lexigram-multimedia-music)
video	VideoConfig	VideoConfig()	Video generation + processing config (lexigram-multimedia-video)
image	ImageConfig	ImageConfig()	Image generation config (lexigram-multimedia-image)
upscale	UpscaleConfig	UpscaleConfig()	Upscale config (lexigram-multimedia-upscale)
interpolate	InterpolationConfig	InterpolationConfig()	Frame interpolation config (lexigram-multimedia-interpolate)
beat	BeatAnalysisConfig	BeatAnalysisConfig()	Beat analysis config (lexigram-multimedia-beat)
storage_path_prefix	str	"multimedia/"	Blob-storage key prefix for generated assets
cache_results	bool	False	Cache accessor generate() results in the cache backend

Per-subsystem fields (notable defaults)

Each nested config carries its backend selector; the full field list lives in each sibling package’s docs.

Subsystem	Field	Type	Default	Description
tts	backend	str	"local-http"	local-http, elevenlabs, openai, chatterbox, kokoro, f5-tts, piper
tts	elevenlabs_voice_id	str | None	null	Required for the elevenlabs backend
music	backend	str	"local-http"	local-http, stability-audio, ace-step, stable-audio-open
video	backend	str	"local-http"	local-http, runway, openai, wan22, cogvideox, svd, comfyui
video	processing.max_concurrent_jobs	int	2	FFmpeg processing concurrency limit
image	backend	str	"local-http"	local-http, stability, openai, comfyui
upscale	backend	str	"real-esrgan"	real-esrgan or hat
interpolate	backend	str	"rife"	RIFE frame-interpolation backend
beat	backend	str	"librosa"	librosa (in-process) or madmom (reference server)
beat	librosa_sample_rate	int	22050	Sample rate for the librosa backend
beat	madmom_base_url	str	"http://localhost:5600"	Madmom reference-server URL

---

Environment Variables

The env var name is the config path in SCREAMING tokens joined by __ under LEX_MULTIMEDIA__. Nested subsystem fields prefix with the subsystem name.

Variable	Description
LEX_MULTIMEDIA__STORAGE_PATH_PREFIX	Blob storage prefix for assets (multimedia/)
LEX_MULTIMEDIA__CACHE_RESULTS	Enable result caching (true/false)
LEX_MULTIMEDIA__TTS__BACKEND	TTS backend name
LEX_MULTIMEDIA__TTS__ELEVENLABS_VOICE_ID	ElevenLabs voice id
LEX_MULTIMEDIA__MUSIC__BACKEND	Music backend name
LEX_MULTIMEDIA__VIDEO__BACKEND	Video backend name
LEX_MULTIMEDIA__IMAGE__BACKEND	Image backend name
LEX_MULTIMEDIA__UPSCALE__BACKEND	Upscale backend name
LEX_MULTIMEDIA__INTERPOLATE__BACKEND	Interpolation backend name
LEX_MULTIMEDIA__BEAT__BACKEND	Beat backend name (librosa/madmom)
LEX_MULTIMEDIA__BEAT__MADMOM_BASE_URL	Madmom server URL

LEX_MULTIMEDIA__TTS__BACKEND=elevenlabs \
LEX_MULTIMEDIA__TTS__ELEVENLABS_VOICE_ID=21m00Tcm4TlvDq8ikWAM \
LEX_MULTIMEDIA__VIDEO__BACKEND=local-http \
LEX_MULTIMEDIA__CACHE_RESULTS=true \
  python -m my_app

A subsystem installed standalone uses its own prefix instead — e.g. LEX_MULTIMEDIA_BEAT__BACKEND (see the beat package’s Configuration doc).

---

Advanced Configuration

Programmatic sub-config overrides

from lexigram.multimedia import MultimediaConfig, MultimediaModule
from lexigram.multimedia.tts.config import TTSConfig
from lexigram.multimedia.beat.config import BeatAnalysisConfig

app.add_module(
    MultimediaModule.configure(
        config=MultimediaConfig(
            tts=TTSConfig(backend="elevenlabs", elevenlabs_voice_id="21m00Tcm4TlvDq8ikWAM"),
            beat=BeatAnalysisConfig(backend="madmom", madmom_base_url="http://10.0.0.5:5600"),
            storage_path_prefix="prod/media/",
        )
    )
)

Standalone subsystem usage

Every subsystem can also be wired on its own with its own module — the umbrella is never required:

from lexigram.multimedia.beat.module import BeatAnalysisModule
from lexigram.multimedia.beat.config import BeatAnalysisConfig

app.add_module(BeatAnalysisModule.configure(config=BeatAnalysisConfig(backend="librosa")))

Using .stub() for tests

# Stubs all installed subsystems; no servers, no network
app.add_module(MultimediaModule.stub())

stub() imports each core subsystem’s .stub() module (AudioTTSModule.stub(...), AudioMusicModule.stub(...), …) and also loads any non-core modules discovered via the lexigram.multimedia.modules entry-point group.

---

Best Practices

• Keep the umbrella config minimal; override only what your deployment differs on.
• Prefer environment variables for secrets and per-environment switches (LEX_MULTIMEDIA__TTS__BACKEND=elevenlabs), YAML for structure.
• Never hardcode API keys in application.yaml — subsystem configs accept *_api_key_secret_name fields that reference secrets-manager entries.
• Enable cache_results only when a cache backend is present; otherwise the flag is inert.
• Give every environment its own storage_path_prefix to avoid cross-env asset collisions.