메뉴
This skill should be used when the user asks to "generate local TTS", "use local text to speech", "use remote TTS over Tailscale", "make speech audio", "add a TTS provider", "use Qwen3-TTS", or needs TTS endpoint setup through local or HivemindOS Tailscale routes.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Use when generating images with Flux.2 on Apple Silicon, working with MLX Swift, implementing text-to-image/image-to-image pipelines, LoRA training, or quantized model inference in Swift.
Use this skill whenever adding, wiring, surfacing, or debugging a new node in an existing ComfyUI Mobile workflow JSON, especially when the user says a node is missing from a workflow they already use. It prevents the common failure mode where the saved workflow file is changed correctly but ComfyUI Mobile still renders an older persisted/open workflow, hides the node as static, or fails to expose custom-node widgets.
Operate the agent-first Hivemind Content Studio for durable first-frame animation ads, stickman ads, static text ads, faceless shorts, clipping, local or cloud media generation, exact voice, assembly, semantic evaluation, approval-gated social posting, experiments, and monetization metrics. Use when an agent must create, resume, inspect, evaluate, publish, or optimize social content through the unified repo rather than invoking Auto Clipper, AI Animation Factory, MoneyPrinterTurbo, MUAPI, Higgsfield, Media Studio, Postiz, or Upload-Post as overlapping standalone pipelines.
Use when setting up or operating AI animation content factories that turn briefs into scripts, frame prompts, motion prompts, voice lines, music briefs, and publishing metadata across tools such as Claude/OpenAI, Midjourney, Runway, ElevenLabs, Suno, Make, and YouTube.
Use the local unpaid Auto Clipper control plane to ingest public creator research URLs, render Podcli clip candidates, require human approval, and schedule only approved clips through self-hosted Postiz.
This skill should be used when the user asks to "generate images", "generate images with ComfyUI", "use local ComfyUI", "use remote ComfyUI", "use Z-Image", "add an image generation provider", or needs image-generation endpoint setup through local or HivemindOS Tailscale routes.
| name | localtts |
| description | This skill should be used when the user asks to "generate local TTS", "use local text to speech", "use remote TTS over Tailscale", "make speech audio", "add a TTS provider", "use Qwen3-TTS", or needs TTS endpoint setup through local or HivemindOS Tailscale routes. |
| version | 0.1.0 |
Generate speech audio through local or HivemindOS Tailnet TTS services. Keep speech providers in this skill as blocks so agents can add new voices and engines without creating one skill per model.
Use one of two modes:
Do not require Tailnet clients to have the TTS repo, model files, LaunchAgent, systemd unit, or localhost port. Do not use client-local 127.0.0.1 for a remote TTS provider.
A local/Tailnet TTS provider should document at least these endpoints. Prefer OpenAI-compatible paths where possible, and add simple aliases only when they make clients easier to write.
Discovery and health
GET /health
GET /v1/models
GET /v1/voices
GET /health: liveness/readiness probe. Include service name, status, loaded model, device/backend, and version when available.GET /v1/models: OpenAI-style model list. Return the model IDs accepted by generation calls.GET /v1/voices: voice discovery. Return stable voice IDs/names, display names, languages, genders/styles if known, and provider-specific metadata needed by clients.Useful optional endpoints
GET /voices
GET /v1/voices/<voice-id>
POST /v1/voices
POST /v1/audio/speech
POST /tts/synthesize
POST /synthesize
POST /api/tts
GET /openapi.json
GET /docs
GET /voices: compatibility alias for clients that do not use /v1/*; should return the same data shape as /v1/voices or clearly document differences.GET /v1/voices/<voice-id>: details for one voice, including language support, sample rate, tags, and whether it requires a reference sample.POST /v1/voices: register a new local cloned voice profile from multipart reference audio. Requires name and ref_audio; accepts language, description, model, ref_text, and x_vector_only_mode.POST /v1/audio/speech: primary OpenAI-style speech generation endpoint. Accept model, input, voice, response_format, and provider extras such as language, instruct, speed, temperature, top_p, max_new_tokens, or seed.POST /tts/synthesize: canonical JSON/base64 compatibility endpoint for clients that cannot handle binary audio responses.POST /synthesize or POST /api/tts: JSON-native compatibility aliases for non-OpenAI clients. Document request/response exactly if the provider supports them.GET /openapi.json and GET /docs: optional API schema and Swagger/ReDoc docs for human/client discovery.Voice endpoint response template
{
"object": "list",
"data": [
{
"id": "<voice-id>",
"name": "<display-name>",
"language": "English",
"languages": ["English"],
"description": "Warm neutral narration voice.",
"styles": ["neutral", "warm", "narration"],
"sample_rate": 24000,
"requires_reference_audio": false,
"provider": "<provider-name>"
}
]
}
Endpoint smoke tests
curl -fsS --max-time 5 http://<host>:<port>/health
curl -fsS --max-time 5 http://<host>:<port>/v1/models
curl -fsS --max-time 5 http://<host>:<port>/v1/voices
curl -fsS --max-time 5 http://<host>:<port>/voices || true
Use this block when documenting an OpenAI-compatible or JSON TTS endpoint reachable through Tailscale.
Client route
http://<magicdns-or-tailnet-ip>:<port>https://<magicdns-or-tailnet-ip>:<port>127.0.0.1 from the client unless the client is also the service host.Tailnet client preflight
curl -fsS --max-time 5 http://<tailnet-host>:<port>/health
curl -fsS --max-time 5 http://<tailnet-host>:<port>/v1/models
curl -fsS --max-time 5 http://<tailnet-host>:<port>/v1/voices
If discovery endpoints fail, diagnose Tailnet reachability or the remote TTS host. Do not substitute unrelated local services on other ports.
OpenAI-style generation example
curl -fsS -X POST http://<tailnet-host>:<port>/v1/audio/speech \
-H 'Content-Type: application/json' \
-d '{
"model": "<model-id>",
"input": "Hello from the HivemindOS Tailnet TTS endpoint.",
"voice": "<voice-name>",
"language": "English",
"instruct": "Speak clearly and warmly.",
"response_format": "wav",
"max_new_tokens": 512
}' \
-o /tmp/tailnet-tts-output.wav
Verify output
ls -lh /tmp/tailnet-tts-output.wav
file /tmp/tailnet-tts-output.wav
Use this block when the current machine runs the TTS service.
Local preflight
curl -fsS --max-time 5 http://127.0.0.1:<port>/health
curl -fsS --max-time 5 http://127.0.0.1:<port>/v1/models
curl -fsS --max-time 5 http://127.0.0.1:<port>/v1/voices
lsof -nP -iTCP:<port> -sTCP:LISTEN || true
Local generation example
curl -fsS -X POST http://127.0.0.1:<port>/v1/audio/speech \
-H 'Content-Type: application/json' \
-d '{
"model": "<model-id>",
"input": "Hello from the local TTS endpoint.",
"voice": "<voice-name>",
"response_format": "wav"
}' \
-o /tmp/local-tts-output.wav
Use this block as a starting point for Qwen3-TTS or a Qwen3-compatible local voice server.
Provider fields to fill in
http://<qwen-tts-host>:<port>http://127.0.0.1:<port><qwen3-tts-model-id><voice-name><path><launchctl/systemd/docker/manual><path or command><mps/cuda/cpu dtype constraints>Service-host diagnostics only
curl -fsS --max-time 5 http://127.0.0.1:<port>/health || true
curl -fsS --max-time 5 http://127.0.0.1:<port>/v1/models || true
curl -fsS --max-time 5 http://127.0.0.1:<port>/v1/voices || true
curl -fsS --max-time 5 http://127.0.0.1:<port>/voices || true
lsof -nP -iTCP:<port> -sTCP:LISTEN || true
# macOS example:
launchctl print gui/$(id -u)/<service-label> || true
# Linux example:
systemctl --user status <service-name> || true
When generating audio for a chat user, save a .wav, .mp3, or .ogg under a stable absolute path and include it in the final response as a native media attachment when the runtime supports it.