Skip to main content

Connect Cloud Inference Backends

Cloud TTS is useful for trials, cross-device use, or setups that do not deploy local models. "Free" usually means free tiers, trial quotas, gifted credits, or limited-time campaigns. Quotas and model availability can change; rely on the provider console.

General Steps

  1. Create an account and API key on the cloud service.
  2. Open Providers in Neiroha and click +.
  3. Select the matching adapter.
  4. Fill Base URL, API Key, and model name when required.
  5. Click Fetch All to fetch models and voices.
  6. Enable the provider.
  7. Click Health Check.
  8. Create a character in Voice Bank and run Quick Test with one short sentence.

Cloud Entry Points

ServiceNeiroha AdapterWhy Try ItDetails
MiMoOpenAI Chat Completions TTSOne key can access TTS, VoiceDesign, and VoiceClone-style models; useful for Chinese and Chinese-English tests.MiMo TTS
Google Gemini TTSGoogle Gemini TTSAI Studio API key; official docs list free-tier limits for Gemini 2.5 Flash Preview TTS.Gemini TTS
Azure SpeechAzure Speech ServiceAzure F0 provides a monthly Neural TTS character quota and stable voice lists.Azure Speech

Free Quota Usage

WorkflowRecommendation
Quick TTSGenerate one sentence at a time to confirm character binding.
Dialogue TTSManually generate 2 to 3 lines before clicking generate all.
Phase TTSSplit 3 to 5 segments first to verify long-text style and cost.
Novel ReaderKeep prefetch low to avoid spending quota immediately.
Video DubbingTest a short subtitle section before generating all cues.

Provider Rate Limits

Cloud failures are often rate-limit failures rather than configuration errors. Set these fields in the provider:

FieldPurpose
Max concurrencyControls simultaneous TTS requests.
RPMRequests per minute; useful for services like Gemini.
TPMTokens per minute; useful for token-limited or token-billed services.
RPDRequests per day; useful for free tiers.

For example, set conservative RPM / TPM / RPD values for gemini-2.5-flash-preview-tts, control Azure F0 by character quota and batch size, and use MiMo Token Plan balance and model consumption rules as the source of truth.