Configuration

Configuration options for lexigram-multimedia-video.

Overview

Configuration lives on two BaseConfig classes:

Class	`config_section`	Purpose
`VideoConfig`	`"multimedia_video"`	Generation subsystem (backend, base URLs, API secrets, ComfyUI knobs)
`VideoProcessingConfig`	`"multimedia_video_processing"`	ffmpeg processing pipeline (nested under `VideoConfig.processing`)

As a subsystem of the lexigram-multimedia umbrella, config is nested under the multimedia: key in application.yaml with prefix LEX_MULTIMEDIA__VIDEO__. Or pass a VideoConfig directly to VideoModule.configure(...).

Zero-config default

VideoModule.configure() → VideoConfig(backend="local-http") — generation against a local server at http://localhost:5004, processing enabled when the ffmpeg binary is on PATH.

Basic Example

multimedia:
  video:
    backend: "comfyui"
    comfyui_base_url: "http://localhost:8188"
    comfyui_checkpoint: "svd_xt_1_1.safetensors"
    processing:
      ffmpeg_binary: "ffmpeg"
      max_concurrent_jobs: 4
      temp_dir: "/var/tmp/lexigram"
      timeout: 300.0

from lexigram.multimedia.video import VideoModule
from lexigram.multimedia.video.config import VideoConfig, VideoProcessingConfig

VideoModule.configure(
    config=VideoConfig(
        backend="comfyui",
        comfyui_base_url="http://localhost:8188",
        processing=VideoProcessingConfig(max_concurrent_jobs=4),
    )
)

Options

`VideoConfig`

Option	Type	Default	Description
`backend`	`Literal["local-http", "runway", "openai", "wan22", "cogvideox", "svd", "comfyui"]`	`"local-http"`	Generation backend selected at registration
`local_http_base_url`	`str`	`"http://localhost:5004"`	Local reference server base URL
`runway_api_key_secret_name`	`str`	`"runway_api_key"`	Secret name resolved via `AsyncSecretStoreProtocol` for Runway
`openai_api_key_secret_name`	`str`	`"openai_api_key"`	Secret name for the OpenAI gateway
`openai_model`	`str`	`"sora-2"`	Default model sent as `payload["model"]` (`VideoRequest.model` overrides)
`openai_base_url`	`str`	`"https://api.openai.com"`	OpenAI-compatible base URL (gateway routing)
`wan22_base_url`	`str`	`"http://localhost:5200"`	Wan2.2 reference server
`cogvideox_base_url`	`str`	`"http://localhost:5201"`	CogVideoX reference server
`svd_base_url`	`str`	`"http://localhost:5202"`	SVD reference server
`comfyui_base_url`	`str`	`"http://localhost:8188"`	ComfyUI server
`comfyui_checkpoint`	`str`	`"svd_xt_1_1.safetensors"`	Checkpoint name filled into the workflow
`comfyui_workflow_path`	`str \| None`	`None`	Custom workflow JSON template (default: bundled `workflows/default_svd.json`)
`comfyui_fps`	`int`	`6`	Frames-per-second placeholder
`comfyui_motion_bucket_id`	`int`	`127`	SVD motion bucket placeholder
`comfyui_poll_interval`	`float`	`1.0`	History poll interval in seconds
`timeout`	`float \| None`	`None`	Overall request timeout; `None` → each backend’s own default
`processing`	`VideoProcessingConfig`	default factory	Nested ffmpeg pipeline config

`VideoProcessingConfig`

Option	Type	Default	Description
`ffmpeg_binary`	`str`	`"ffmpeg"`	ffmpeg executable (`ffprobe` is derived from its name)
`max_concurrent_jobs`	`int`	`2`	Semaphore cap on concurrent ffmpeg jobs
`temp_dir`	`str \| None`	`None`	Workdir root for materialized inputs/outputs
`timeout`	`float`	`300.0`	Per-job ffmpeg timeout in seconds (kills the process)

Backend constructor mapping (`VideoGenerationProvider.register()`)

`backend`	Class	Key construction args
`"local-http"`	`LocalHttpVideoProvider`	`base_url=local_http_base_url`
`"runway"`	`RunwayVideoProvider`	`api_key` from secret `runway_api_key_secret_name`
`"openai"`	`OpenAIVideoProvider`	`api_key` from secret `openai_api_key_secret_name`, `model=openai_model`, `base_url=openai_base_url`
`"wan22"`	`Wan22VideoProvider`	`base_url=wan22_base_url`
`"cogvideox"`	`CogVideoXVideoProvider`	`base_url=cogvideox_base_url`
`"svd"`	`SVDVideoProvider`	`base_url=svd_base_url`
`"comfyui"`	`ComfyUiVideoProvider`	`base_url`, `checkpoint`, `workflow_path`, `fps`, `motion_bucket_id`, `poll_interval`
anything else	raises `ProviderNotInstalledError`	—

timeout is forwarded only when set (_TimeoutKwargs); otherwise provider defaults apply: local-http/runway/openai 60.0, wan22/cogvideox 180.0, svd 120.0, comfyui 120.0. All constructors also receive optional retry/circuit_breaker.

Environment Variables

Variable	Description
`LEX_MULTIMEDIA__VIDEO__BACKEND`	Backend selection
`LEX_MULTIMEDIA__VIDEO__LOCAL_HTTP_BASE_URL`	Local server URL
`LEX_MULTIMEDIA__VIDEO__RUNWAY_API_KEY_SECRET_NAME`	Secret name for Runway key
`LEX_MULTIMEDIA__VIDEO__OPENAI_API_KEY_SECRET_NAME`	Secret name for OpenAI key
`LEX_MULTIMEDIA__VIDEO__OPENAI_MODEL`	Default OpenAI gateway model
`LEX_MULTIMEDIA__VIDEO__OPENAI_BASE_URL`	OpenAI-compatible base URL
`LEX_MULTIMEDIA__VIDEO__WAN22_BASE_URL`	Wan2.2 server URL
`LEX_MULTIMEDIA__VIDEO__COGVIDEOX_BASE_URL`	CogVideoX server URL
`LEX_MULTIMEDIA__VIDEO__SVD_BASE_URL`	SVD server URL
`LEX_MULTIMEDIA__VIDEO__COMFYUI_BASE_URL`	ComfyUI URL
`LEX_MULTIMEDIA__VIDEO__COMFYUI_CHECKPOINT`	ComfyUI checkpoint
`LEX_MULTIMEDIA__VIDEO__COMFYUI_WORKFLOW_PATH`	Custom workflow JSON path
`LEX_MULTIMEDIA__VIDEO__COMFYUI_FPS`	Workflow fps
`LEX_MULTIMEDIA__VIDEO__COMFYUI_MOTION_BUCKET_ID`	Workflow motion bucket
`LEX_MULTIMEDIA__VIDEO__COMFYUI_POLL_INTERVAL`	Poll interval seconds
`LEX_MULTIMEDIA__VIDEO__TIMEOUT`	Request timeout
`LEX_MULTIMEDIA__VIDEO__PROCESSING__FFMPEG_BINARY`	ffmpeg binary
`LEX_MULTIMEDIA__VIDEO__PROCESSING__MAX_CONCURRENT_JOBS`	Concurrency cap
`LEX_MULTIMEDIA__VIDEO__PROCESSING__TEMP_DIR`	Workdir root
`LEX_MULTIMEDIA__VIDEO__PROCESSING__TIMEOUT`	Per-job ffmpeg timeout

export LEX_PROFILE=production
export LEX_MULTIMEDIA__VIDEO__BACKEND=runway
export LEX_MULTIMEDIA__VIDEO__PROCESSING__MAX_CONCURRENT_JOBS=4

Advanced Configuration

Secrets for hosted backends

Keys are resolved by name at registration:

# Secrets stored under name "runway_api_key" → RunwayVideoProvider(api_key=...)
# Secrets stored under name "openai_api_key" → OpenAIVideoProvider(api_key=...)

No key → api_key="" and _credential_resolved=False; the health check then reports DEGRADED for that provider. A 401 at call time raises VideoGenerationAuthenticationError.

Testing with `stub()`

module = VideoModule.stub()
# VideoConfig(backend="local-http"), real module — safe for boot smoke tests.

Resolving health per backend

HTTP backends check GET /health (ComfyUI uses GET /system_stats) — 200 → HEALTHY, else DEGRADED.
Runway/OpenAI: HEALTHY only when _credential_resolved (an API key was found).

Entry-point discovery

The umbrella imports this subsystem via lexigram.multimedia.subsystems (video) and lexigram.multimedia.modules (video); the compound config there nests under multimedia: / video:.

Best Practices

✅ Keep config minimal — backend + one base URL per deployment.
✅ Store API keys in the secrets backend; reference them by name — never commit keys.
✅ Give processing.timeout headroom over your longest encode.
✅ Isolate each reference server in its own venv (torch is heavy).
❌ Don’t rely on timeout: null semantics — set it explicitly if your generation SDK is slow.
❌ Don’t set comfyui_workflow_path to an unreadable/unreachable path — the provider reads it at request time.