This ComfyUI workflow turns written scripts into expressive speech using ElevenLabs’ Eleven v4 model. It’s designed around two clear paths: use a preset voice from your ElevenLabs library or instantly clone a voice from an audio sample. The core generation happens in the ElevenLabsTextToSpeech node, which reads your script (written in a Note/MarkdownNote) and applies performance directions you embed as inline tags like [whispers], [excited], [laughs], [sighs], [sarcastic], and [crying]. SaveAudioAdvanced then writes a single audio file to disk.

Technically, Option 1 (Use Preset Voice) routes a selection from ElevenLabsVoiceSelector into ElevenLabsTextToSpeech along with your script and the Eleven v4 model choice. Option 2 (Create a Voice Clone) feeds audio from LoadAudio or RecordAudio into ElevenLabsInstantVoiceClone, which returns a temporary voice you can target in the TTS node. Only one option group should be enabled at a time (select the group and press Ctrl-B to un-bypass it). Because Eleven v4 treats the prompt as a performance script, the workflow lets you direct tone and delivery directly in text while keeping the node graph compact and easy to iterate for narration, audiobooks, and character voiceovers.