qwen3 / customVoice

The fastest path to speech in this whole pack. CivitaiTextToSpeechVllmOmniQwen3CustomVoice is the "customVoice" mode of Qwen3 TTS on Civitai's cloud: instead of feeding it reference audio, you pick from nine built-in named speakers - aiden, dylan, eric, ono_anna, ryan, serena, sohee, uncle_fu, vivian - type your text, and get an AUDIO output back. No reference clips, no transcripts, no local model. It's the node you reach for when you need a voice now and don't care who it belongs to.

It lives in Civitai/Audio/qwen3, part of Civitai's official ComfyUI pack. Compared to its siblings - Base (clone from a reference clip) and VoiceDesign (describe a voice in prose) - CustomVoice is the low-friction middle: more variety than a single default voice, none of the cloning workflow's fiddliness. Good for throwaway narration, multi-character demo dialogue where each line needs a distinct voice, or just poking around to hear what the engine can do.

How it works

The node submits a textToSpeech workflow with operation: customVoice to Civitai's Orchestration API. The speaker dropdown maps to a built-in voice the engine already knows - no upload, no embedding step. The pack long-polls the job and returns the generated audio as a native AUDIO tensor on audio_blob, with model_type and speaker reported back so you can see exactly which voice actually spoke.

The inputs that matter

• text (required) - what the voice should say.
• speaker (required) - one of the nine built-in names. This is the whole personality of the output; try a couple and pick the one that fits the vibe.
• instruct - an optional style instruction like "speak slowly and clearly" or "energetic, like a game-show host." Cheap way to bend a voice without changing it.
• language - target language, empty means auto.
• max_new_tokens - optional cap on generated tokens; 0 = no cap.

Outputs: audio_blob (AUDIO), model_type, speaker, plus the standard workflow_id and raw_json for cost tracking.

Installing it

This is one of ~160 nodes in Civitai Comfy Nodes, Civitai's official pack for their Orchestration API. Install once, all nodes appear under the Civitai menu:

• ComfyUI Manager: Manager → Custom Nodes Manager → search Civitai Comfy Nodes → Install, then restart.
• CLI: comfy node registry-install civitai-comfy-nodes
• Source: cd ComfyUI/custom_nodes && git clone https://github.com/civitai/civitai-comfy-nodes.git && pip install -r civitai-comfy-nodes/requirements.txt (only requests).

You need a Civitai account with Buzz and credentials - a Civitai Auth node, CIVITAI_API_TOKEN for headless/remote, or a stored key from the Civitai sidebar.

Where people get burned

• No credentials → immediate error. Set CIVITAI_API_TOKEN and restart ComfyUI; the browser-login path is a non-starter on headless installs.
• Speaker names are opaque. The dropdown doesn't tell you who sounds like what, and the API gives you no preview. Budget one cheap test clip per name before committing to a voice for a project.
• Paid and cloud-only. Every clip burns Buzz and your text goes to Civitai's servers, where platform filters apply. This is not the node for content the site would refuse.
• Early preview. The pack README warns of breaking changes without notice, and early community reports flagged bugs and slow generations. Long clips can hit the default 30-minute timeout - raise it via the Auth node or CIVITAI_COMFY_TIMEOUT.

If the speaker list has what you need, this is the least-effort TTS node in the pack. Type, pick, run.