Configuration

Configuration options for lexigram-multimedia-upscale.

---

Overview

Configuration lives on UpscaleConfig (a lexigram.config.BaseConfig subclass). As a subsystem of the lexigram-multimedia umbrella, it is nested under the multimedia: key in application.yaml with environment-variable prefix LEX_MULTIMEDIA__UPSCALE__. Alternatively, pass an UpscaleConfig directly to UpscaleModule.configure(...).

Config class	config_section	Provider
UpscaleConfig	"multimedia_upscale"	consumed by UpscaleGenerationProvider (provider config_key = "multimedia_upscale")

Zero-config default

Calling UpscaleModule.configure() with no arguments produces UpscaleConfig(backend="real-esrgan") — super-resolution against a reference server at http://localhost:5400.

---

Basic Example

multimedia:
  upscale:
    backend: "hat"                     # real-esrgan | hat
    real_esrgan_base_url: "http://localhost:5400"
    hat_base_url: "http://localhost:5401"
    timeout: 30.0

from lexigram.multimedia.upscale import UpscaleModule
from lexigram.multimedia.upscale.config import UpscaleConfig

UpscaleModule.configure(
    config=UpscaleConfig(backend="hat", hat_base_url="http://10.0.0.5:5401")
)

---

Options

UpscaleConfig

Option	Type	Default	Description
backend	Literal["real-esrgan", "hat"]	"real-esrgan"	Which super-resolution backend to build at registration time
real_esrgan_base_url	str	"http://localhost:5400"	Base URL of the Real-ESRGAN reference server
hat_base_url	str	"http://localhost:5401"	Base URL of the HAT reference server
timeout	float	30.0	Per-request HTTP timeout in seconds (aiohttp ClientTimeout.total)

The upscale factor is not config-state: pass it per request via UpscaleRequest.scale_factor (Literal[2, 4], default 4).

Backend constructor mapping

The provider forwards config into the backend constructors in UpscaleGenerationProvider.register():

backend	Class constructed	Args
"real-esrgan"	RealEsrganUpscaleProvider	base_url=real_esrgan_base_url, timeout, retry, circuit_breaker
"hat"	HatUpscaleProvider	base_url=hat_base_url, timeout, retry, circuit_breaker
anything else	raises ProviderNotInstalledError	—

retry / circuit_breaker come from optional RetryPolicyProtocol / CircuitBreakerProtocol bindings in the container.

---

Environment Variables

Variable	Description
LEX_MULTIMEDIA__UPSCALE__BACKEND	Override backend
LEX_MULTIMEDIA__UPSCALE__REAL_ESRGAN_BASE_URL	Override real_esrgan_base_url
LEX_MULTIMEDIA__UPSCALE__HAT_BASE_URL	Override hat_base_url
LEX_MULTIMEDIA__UPSCALE__TIMEOUT	Override timeout

export LEX_PROFILE=production
export LEX_MULTIMEDIA__UPSCALE__BACKEND=hat
export LEX_MULTIMEDIA__UPSCALE__HAT_BASE_URL=http://10.0.0.5:5401

---

Advanced Configuration

Testing with stub()

module = UpscaleModule.stub()
# Equivalent to configure(config=UpscaleConfig(backend="real-esrgan"))
# — useful in test suites; the module is real, the default backend is pinned.

Wiring through an umbrella module

@module(imports=[UpscaleModule.configure()])
class AppModule(Module):
    pass

The umbrella lexigram-multimedia orchestrator may also import this subsystem by entry point lexigram.multimedia.subsystems (upscale) or lexigram.multimedia.modules (upscale) — no config needed in that case beyond the multimedia: upscale: YAML section.

Optional cross-package wiring

VideoUpscaleService only registers when a VideoProcessor is resolvable in the container (provided by lexigram-multimedia-video when ffmpeg is on PATH). No upscale-side flag controls this — presence of the processor is the switch.

---

Best Practices

• ✅ Prefer environment variables / YAML for backend choice; keep secrets out of config entirely.
• ✅ Use per-request scale_factor rather than forking config per scale.
• ✅ Point *_base_url at machines that run lexigram-upscale-*-serve.
• ❌ Don’t hardcode model weights or server code in the application — the reference servers are separate processes.
• ❌ Don’t expect backend="real-esrgan" to work without a server listening on the configured port.