TTS WebUI Vall-E-X

Vall-E-X is the open reimplementation of Microsoft's Vall-E, and in a real sense it's the reason voice-cloning TTS became a normal thing. Released back in 2023, it proved you could clone a voice from a few seconds of audio and generate speech in that voice across many languages - no training required. It's dated now, and the newer models in this pack generally beat it, but it's still here because it works, it's multilingual, and it's lighter than the 2025-era models. If you want to know what all the fuss was about, this is the node.

How it works

Same client-server plumbing as the rest of the pack: the node POSTs an OpenAI-style body to http://127.0.0.1:7778/v1/audio/speech with model: "vall-e-x", and the TTS WebUI server does the synthesis. Vall-E-X is a codec-based approach - it tokenizes audio with EnCodec and generates speech tokens conditioned on the reference voice. The reference voice itself is configured on the server side, so this node mostly supplies the transcript and the language.

The inputs that matter

• prompt - the transcript of the reference voice clip. Vall-E-X aligns the reference by knowing what it says, so this should match the server's reference audio. This is the input people most often leave empty and then wonder why cloning is poor.
• language - defaults to English. Vall-E-X is genuinely multilingual; you can generate in other languages using the cloned voice.
• accent - defaults to no-accent. The knobs here are about how much of the source accent carries over.
• mode - defaults to short. The other common value is long-form mode, which handles longer generations better.

Outputs

audio (AUDIO) into a preview/save node, wav_path (STRING) when also_save_wav is on, and metadata_json (STRING) with the request details.

Installing it

ComfyUI Manager → search "TTS WebUI API nodes for ComfyUI", or:

cd ComfyUI/custom_nodes
git clone https://github.com/rsxdalv/ComfyUI-TTS-Webui

then restart. The pack only needs requests, numpy, torch. Vall-E-X's model files download into the TTS WebUI server on first load - it's an older, chunkier model, so budget some patience for that first request.

When it breaks

• Cloning sounds weak or off: check that the server has a reference voice loaded and that prompt matches its transcript. Vall-E-X lives or dies on that pairing.
• Connection refused: server not running on 127.0.0.1:7778, or api_base points at the wrong host (the node appends /v1/audio/speech itself).
• First call times out: model download/load server-side. Raise timeout_sec (max 600).
• Long text degrades: switch mode to long-form rather than fighting it.

The honest verdict: unless you specifically need Vall-E-X's multilingual cloning or its particular character, you're probably better off with F5-TTS or Chatterbox for quality. But this node is the historical record, and for a quick clone in an unusual language it's still got a niche.