Kokoro TTS v1

If you just want text turned into a spoken voice and don't care about mixing two voices together, this is the node you actually want. No blend modes, no voice-shape files, no dial to tune - type text, pick a voice, get audio. Everything else in this pack (and it's a pack that's mostly about blending) builds on top of what this node already does.

What it's running on

Kokoro is a real, well-regarded open TTS model - tiny by TTS standards (around 82M parameters) but punching way above its size, which is exactly why it's become one of the go-to picks for local text-to-speech in the ComfyUI crowd. This particular node doesn't run the raw PyTorch weights, though - it's built on the ONNX port of Kokoro (kokoro-onnx), which is lighter to deploy: no training-stack dependencies, just a couple of model files and an ONNX runtime. DJZ-KokoroTTS is one of a handful of Kokoro wrappers floating around for ComfyUI; this one's by MushroomFleet (Drift Johnson), who's better known for the much larger DJZ-Nodes suite. Think of this pack as a focused spinoff, not his flagship.

Inputs and outputs

Three inputs, and you'll only ever touch two of them:

• text (STRING) - what gets spoken. Defaults to "Hello, world!" so you know it's wired up before you replace it.
• voice (enum, 46 choices) - the built-in Kokoro voice presets, named by accent and gender: af_* and am_* are American female/male, bf_*/bm_* are British. Pick one and that's your speaker.
• speed (FLOAT, 0.5–2, default 1) - straightforward playback rate, not pitch.

One output: audio (AUDIO). Wire it into a Save Audio or Preview Audio node to actually hear or export it, or further downstream if you're muxing narration into a video.

Installing it

Two ways in, same as most custom node packs:

• ComfyUI Manager - search "DJZ-KokoroTTS", install, restart.
• Manual - cd ComfyUI/custom_nodes && git clone https://github.com/MushroomFleet/DJZ-KokoroTTS, then restart.

If you're on a Windows portable build, there's a dedicated install-portable.bat in the repo that wires up dependencies and paths for you - close ComfyUI first, then run it. Running it against a live instance is asking for file-lock errors.

The part people skip: this node needs actual model files, and the repo doesn't bundle them. Two files need to land in /comfyui/models/kokoro/. The README gives you a direct link for one of them - the ONNX weights themselves, kokoro.onnx, from the taylorchu/kokoro-onnx v0.2.0 release - but it's vaguer about the second file, pointing you at the repo's own models.json for the exact filename it expects. Check that file before you assume you're done.

Where this actually goes wrong

By the README's own admission, the model files are the number one failure mode - the node loads fine, then errors out or the voice list comes up empty because something isn't sitting in exactly /comfyui/models/kokoro/. Not a subfolder, not renamed, exactly that path.

The second thing worth knowing, and it's not specific to this pack: every custom node in ComfyUI installs its own requirements.txt into the same shared Python environment, with no isolation between them. If you've got a big node collection already, a version conflict between this pack's dependencies and something else's is a real possibility, and it tends to show up as an unrelated-looking import error rather than a clean message pointing at Kokoro. If the node fails to load at all (not just fails to generate audio), that's the first thing to check - the ComfyUI console log on startup will name the actual missing or conflicting package.

Beyond that: this is a young, fairly obscure pack - don't expect a big Reddit thread to bail you out if you hit a weird edge case. The repo's own issues page is your best bet.