franckyb/comfyui-dramabox

ComfyUI-DramaBox adds text-to-speech nodes to ComfyUI, a visual, node-based tool people use to build AI image and audio generation workflows. It is a port of DramaBox, a speech synthesis system built by ResembleAI, so it can be used inside ComfyUI like any other building block. DramaBox is described as an expressive text-to-speech system built on LTX-2.3, an audio diffusion transformer, which is a type of AI model that generates audio step by step, similar to how image diffusion models generate images. The extension gives you two nodes to drag into your workflow. The first, DramaBox TTS, takes a text prompt and turns it into spoken audio. You can optionally give it a short audio clip of a real voice as a reference, and it will try to match that voice's style. The second node, DramaBox Options, lets you adjust generation settings such as the number of generation steps, CFG scale, and audio duration. On first use, the extension automatically downloads the required model weights. It also supports LoRA files, small trained add-ons that bake a particular voice style into the model. LoRAs are specific to this model and cannot be reused with other ComfyUI nodes such as LTX Video. You can train your own voice LoRAs with a companion tool called Voice Clone Studio, DramaBox Edition, which transcribes and splits long audio clips into smaller pieces ready for training, then produces a file you drop into ComfyUI's loras folder. Using a LoRA together with a voice reference clip reinforces a specific voice style. Installation involves cloning the repository into ComfyUI's custom_nodes folder, activating the ComfyUI virtual environment, and running a pip install of the requirements. The README also documents a workaround for a known build issue with the bitsandbytes library on newer CUDA toolkit versions, which requires compiling it from source. The project is written in Python.