wtf is trdhiggsv3?

TrdHiggsV3 is a custom node for ComfyUI that converts text into speech using a cloned voice, running entirely on your own machine with no cloud uploads. ComfyUI is a visual, node based tool commonly used for building AI workflows. This project was built after the author, a freelance developer, took on a job converting text to speech in a client's own voice, and found that most existing text to speech models were trained on Brazilian Portuguese rather than the European Portuguese the client actually spoke, so the accent and rhythm came out wrong. The node connects to a locally running speech model called Higgs Audio v3, served through a backend called sglang-omni, and sends it text to turn into audio. Because feeding a speech model a huge block of text at once tends to degrade quality, the node automatically splits long text into chunks of about 250 characters, breaking at sentence boundaries, generates audio for each chunk separately, then stitches the resulting audio clips back together into one continuous file. Voice cloning works by providing a short reference recording of the target speaker's voice, roughly five to fifteen seconds long, along with a written transcript of exactly what was said in that recording. The model uses this reference to match the accent, rhythm, and tone of the generated speech, which the author says makes a real difference for European Portuguese specifically. To install it, you clone the repository into ComfyUI's custom nodes folder and install its Python dependencies, then the node appears in ComfyUI under the Audio and TTS category. Using it requires the Higgs Audio v3 server to already be running locally before you use the node in a workflow. The author lists several known limitations honestly, including no timeout if the server hangs, no automatic retry after a network error, and the fact that very long single sentences are not split further even if they exceed what the model can handle well. The project is released under the MIT license.