Scenema Audio Released as ComfyUI Node Running on 8GB VRAM
Scenema Audio has been released as a native ComfyUI custom node, quantized to run locally on consumer GPUs with at least 8GB of VRAM. The update features expressive text-to-speech, zero-shot voice cloning, and inline stage directions using bracket cues instead of the previous XML format. This release democratizes advanced generative audio by allowing creators to run heavy transformer-based voice synthesis locally on consumer-grade hardware. Integrating it into ComfyUI enables seamless combination with other generative AI workflows like image and video creation. The model uses Gemma 3 12B as its text encoder, requiring users to accept its HuggingFace license and set an `HF_TOKEN` before the first run, which downloads approximately 30GB of weights. Because it is a diffusion-based model, it may occasionally produce repetition or gibberish, making it best suited for a generate-and-trim post-editing workflow.
## BACKGROUND
ComfyUI is a popular node-based graphical user interface used to build modular workflows for generative AI models. Model quantization is a technique that reduces the precision of a model's parameters to shrink its memory footprint, allowing large models to run on smaller GPUs. Zero-shot voice cloning allows a system to replicate a specific voice using only a short reference audio clip without any additional training.