ElevenLabs Voice Design

No reference audio, no sample to upload, no cloning at all. ElevenLabs Voice Design generates a voice from nothing but a description - "a deep, authoritative male voice with a slight British accent" - and immediately plays it back speaking your own sample sentence. It's the fastest way in this pack to audition a character voice, and honestly one of the most fun nodes to futz with.

How it works

The node POSTs your voice_description, the sample_text, and the structural attributes to ElevenLabs' /v1/voice-generation/generate-voice endpoint. That's a voice generation preview - the API synthesizes a brand-new voice on the fly and renders your sample text with it. You get back two things:

• preview_audio (AUDIO) - the sample text spoken in the designed voice.
• voice_info (STRING) - a formatted summary of what you asked for plus a status message.

Here's the catch, and the node's own voice_info text says it outright: the designed voice is not saved to your account. It's a preview. If you love it, you can't grab a voice ID from this node and start using it in TTS - there's no voice ID output at all. To make it permanent you'd need to go through ElevenLabs' web UI or build a proper clone from the preview audio. So treat this as an audition tool, not a voice factory.

Inputs

• voice_description - free-text description. This does the heavy lifting; be specific.
• sample_text - the sentence it speaks. Use the actual line you care about, because that's what you'll be judging.
• gender - male, female, or neutral.
• age - young, middle_aged, or old.
• accent - american, british, australian, indian, african, or neutral.
• accent_strength - 0.5 to 2.0, default 1.0. This is the dial to play with; crank it to 2.0 and the accent gets thick.

Install

Same pack, same drill - Manager search "Elevenlabs-ComfyUI", or:

cd ComfyUI/custom_nodes
git clone https://github.com/karthikg-09/Elevenlabs-ComfyUI.git
cd Elevenlabs-ComfyUI && pip install -r requirements.txt

Restart, then Add Node → ElevenLabs. Only real dependency beyond ComfyUI's stock torch is requests.

The honest take

This is a design-fast, keep-nothing node. Its real value is in character pipelines where you want to iterate over a dozen voice ideas before committing: describe, preview, discard, describe again - each pass is one cheap API call. Where it gets you nowhere is persistence, and the README's claim that it outputs a voice_id you can reuse in TTS is simply not what the node ships. Don't build a workflow that expects a saved voice out of this; build one that expects a decision.

One workflow that actually works: design a voice you like, then use the Voice Clone node with the preview audio (or, better, a cleaner long-form sample) to turn the concept into a permanent account voice for TTS. Design for the idea, clone for the asset.