Available is not active

A downloaded model folder proves that model files exist. It does not prove that the matching Python backend can import, that VoiceStudio selected that backend, or that new synthesis requests route through it.

VoiceStudio reports those facts separately. Read them in the order below when an engine looks installed but unavailable, or when an older model remains in memory after another engine becomes active.

Model files or installer stateCopy Anchor Link

An installer can finish, or a Hugging Face cache directory can exist, while a backend dependency or configuration value is still missing. Treat downloaded files as one prerequisite, not as a readiness verdict.

This distinction matters for engines with more than one layer. In v0.5.1, CosyVoice first checks whether its Python backend imports. It resolves the model directory later, when the engine attempts its first load. A model cache alone therefore cannot prove that CosyVoice is ready to synthesize. The tagged CosyVoice adapter shows both checks.

VoxCPM2 exposes the same boundary from the other direction. The Models pane marks openbmb/VoxCPM2 installed from its complete Hugging Face cache snapshot. The engine probe separately imports the voxcpm Python package. In v0.5.1, the one-click sidecar provisioner has no VoxCPM2 entry, so downloading the weights cannot make a missing runtime import available. Keep the weights, open Why unavailable, and record the exact reason before changing a Python environment. The tagged model inventory, VoxCPM2 probe, and sidecar specifications show the three separate checks.

AvailableCopy Anchor Link

The Available badge comes from that backend's availability probe. Each backend defines its own probe, so the exact requirement may be a Python package, an external process, a configured path, or a supported host.

If an engine is Unavailable, open Why unavailable before reinstalling anything. The row can expose the backend's reason, install hint, last error, or a setup line. The tagged backend inventory defines those fields.

Available means the readiness probe passed. It is not proof that a model has already loaded or that a synthesis request has succeeded.

ActiveCopy Anchor Link

Active identifies the backend selected to receive new work for its family. It is separate from availability and memory residency.

VoiceStudio resolves the active TTS backend from an OMNIVOICE_TTS_BACKEND environment override first, then the saved UI choice, then the default. The tagged preference resolver keeps that precedence explicit. When the environment override is present, changing the picker cannot silently replace it.

If the active badge does not follow the UI selection, check how VoiceStudio was launched and whether that environment variable is set in the launch context. Do not remove an override until you know why it was added.

Routing deviceCopy Anchor Link

The routing badge names the effective device for that engine on the current machine. GPU compatibility and active selection are different facts: an engine can be available but use the CPU, or be unavailable before routing matters.

Read the routing reason when VoiceStudio reports CPU fallback or an unavailable device path. Do not infer acceleration from installed GPU drivers or model files alone.

Loaded or in memoryCopy Anchor Link

Loaded models describes memory residency. A model may still occupy RAM or VRAM while another available engine is active. VoiceStudio can label this state as in memory and can mark a loaded model as not active. The tagged quick switch shows these labels separately.

The Loaded models list answers a memory question. The Active badge answers a routing question. Do not use one as proof of the other.

If an old model is consuming memory, use its Unload control or Unload all and flush. Unloading memory does not install dependencies or change an environment override.

Read the engine row in this orderCopy Anchor Link

1. Check Available or Unavailable.
2. If unavailable, open Why unavailable and keep the exact reason.
3. Check which engine carries the Active marker.
4. Read the routing badge for the effective device on this machine.
5. Use Loaded models only to inspect or release resident memory.

For a support report, include the VoiceStudio version, operating system, engine ID, exact availability reason, active-engine marker, and whether an environment override is set. If first load fails after the engine becomes available, include that load error separately. These details distinguish setup, selection, routing, and runtime failures without asking another person to infer them from one screenshot.

For help choosing an engine after its state is clear, use the task-first engine guide. For error payloads and support details, read Errors and request IDs.