Kokoro TTS v5 (Voice Shape Blender)

This is the node for people who've already gone one level deep on voice blending and want to go a second. KokoroTTS_v5 - the pack calls it the "Voice Shape Blender" - takes the same blend engine as KokoroTTS_v3, but instead of picking two of Kokoro's 46 stock presets, both inputs are custom VOICE shapes you built and saved earlier. In other words: it blends blends.

Why you'd want this

The pack's normal flow is preset → blend → save → reload → generate. That gets you one custom voice. v5 exists for when one custom voice isn't the end of the road - say you've built two different blended voices for two different characters and want a third voice that sits between them, or you're iterating toward a specific sound and want to keep folding new material into what you already have. It's the same slerp-vs-weighted-average machinery as v3, just pointed at custom voices instead of raw presets. This is a genuinely niche use case - most people building one custom voice and sticking with it will never touch this node - but if you're doing serious voice design in this pack, it's the tool that lets you actually iterate rather than start over from two presets every time.

Inputs and outputs

• text (STRING, default "Hello, world!") - what gets spoken.
• voice1 / voice2 (VOICE) - two custom voice shapes, wired in from KokoroTTS_LoadVoice_v1 (or however else you're producing a VOICE value upstream). Not enums - you can't pick a stock preset here.
• speed1 / speed2 (FLOAT, 0.5–2, default 1) - playback speed per source voice.
• blend_modes (enum: default, slerp) - weighted-sum average or spherical interpolation.
• slerp_t (FLOAT, 0–1, default 0.5, step 0.01) - the mix ratio between voice1 and voice2. Note the finer step here (0.01) compared to v3's 0.1 - this node is built for fine-tuning an already-close blend, not broad strokes.

One output: audio (AUDIO).

Installing it

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

Windows portable users get an install-portable.bat in the repo - close ComfyUI before running it, since it needs to write into a live install.

You'll also need two model files in /comfyui/models/kokoro/. The README links directly to the ONNX weights (kokoro.onnx, from taylorchu/kokoro-onnx's v0.2.0 release), but doesn't name the second file precisely - check the repo's models.json for what it actually expects there.

Where people get stuck

The obvious one first: this node needs two already-built VOICE shapes, not presets, so you can't use it in isolation. You need KokoroTTS_SaveVoice_v1 and KokoroTTS_LoadVoice_v1 run twice each (once per custom voice) before v5 has anything valid to blend. If you only have one saved voice so far, this isn't the node you want yet - go back to v3 or v4.

Past that, it inherits the whole pack's usual failure modes: model files not exactly in /comfyui/models/kokoro/ is the most common one per the README's own troubleshooting notes, and - like any custom node pack - its requirements.txt shares ComfyUI's Python environment with everything else you've installed, so a version conflict elsewhere can break it on load with an error that won't obviously mention Kokoro. Check the startup console log first if the node refuses to load at all.

This is a low-traffic, advanced node in a fairly obscure pack - don't expect much community troubleshooting out there yet if something goes sideways.