Generate Speech [BETA] (Venice)

Text-to-speech inside ComfyUI usually means dragging in a local TTS model and babysitting its dependencies. Generate Speech [BETA] (Venice) skips all of that: type your line, pick a voice, and venice.ai's tts-kokoro model returns a ready-to-use AUDIO tensor. No model download, no GPU grind - but read the label before you plan a production pipeline around it.

This node is beta on two levels. Venice itself had TTS marked as beta as of mid-2025, and the pack author has flagged the node EXPERIMENTAL in code. The practical consequence: ComfyUI's node search won't show it by default. You have to enable Settings > Show experimental nodes in search (it's under the Node settings) before "Generate Speech" appears at all. That trips up a surprising number of people on first install - you'll think the pack didn't load, when really the node is just hiding.

How it works

The node POSTs your text to https://api.venice.ai/api/v1/audio/speech, saves the returned audio to a temp file, decodes it with torchaudio, and hands back a {waveform, sample_rate} dict - the standard ComfyUI AUDIO format. Wire that into whatever audio output or save node you normally use and you're done.

The inputs

• model - default tts-kokoro, currently the only TTS model Venice offers. Populated from your account's model list, so it needs a working API key.
• input - the text to speak, maximum 4096 characters. The node rejects both empty input and anything over the cap.
• response_format - mp3 (default, lossy but universally supported), wav (lossless), or pcm (raw). The code comments hint the author tried more formats and noped out, so these three are the stable ones.
• speed - 0.25 to 4.0, default 1.0. And here's a small comedy: the tooltip on this field is a copy-paste leftover from the image-enhance node, so it prattles on about "text to image style" prompts. Ignore it - it's the playback speed.
• voice - a dropdown of voices from your account, like af_sky - tts-kokoro. The node strips the model suffix before sending, so "af_sky - tts-kokoro" becomes just "af_sky" on the wire.

Output

audio - an AUDIO dict compatible with ComfyUI's audio nodes, ready to save or play.

Installing it

Part of the ComfyUI-Venice-API pack - one install covers all seven nodes:

cd ComfyUI/custom_nodes
git clone https://github.com/DraconicDragon/ComfyUI-Venice-API.git
pip install -r ComfyUI-Venice-API/requirements.txt

Or via ComfyUI Manager (search "ComfyUI-Venice-API"). Restart, set your key in Settings > VeniceAI > API Key, and flip on experimental nodes in search. One dependency worth knowing about: the node imports torchaudio, which isn't in the pack's requirements.txt - it normally comes bundled with ComfyUI's Python environment, but if you get No module named torchaudio, pip install torchaudio fixes it. Everything else runs remotely, so the only real "cost" is the per-character API usage.