ReleasesNSFWPlatform

DramaBox brings expressive prosody to open-weight voice synthesis

Resemble AI released DramaBox, an expressive voice synthesis model built on LTX 2.3 latent diffusion, with open weights, inference code, and ComfyUI integration.

May 13, 2026

DramaBox brings expressive prosody to open-weight voice synthesis

Resemble AI released DramaBox this week, an expressive voice synthesis model built on Lightricks' LTX 2.3 latent diffusion architecture. The model weights, inference code, and a live HuggingFace Space are all available under Apache 2.0 license, making it open for commercial use. A ComfyUI integration appeared within hours of launch, letting users wire the model into visual workflows.

More in Releases

Watch: DramaBox brings expressive prosody to open-weight voice synthesis

DramaBox targets prosodic range — the demo clips show varied intonation, pacing, and emotional coloring across different prompts. Resemble AI positions it as a step beyond flat text-to-speech, aiming at audiobook narration, character voices for games, and conversational agents that need vocal performance rather than robotic delivery. The HuggingFace Space lets users test short prompts in-browser without local setup.

Architecture and deployment

The choice to build on LTX 2.3 rather than a purpose-built audio diffusion framework is notable — it suggests the latent structure generalizes well across modalities. The GitHub repo includes inference scripts and environment setup for practitioners running the model on local hardware. GPU memory requirements and context length limits aren't specified in the initial release notes, so early adopters will need to benchmark their own rigs.

Whether DramaBox's expressiveness holds up under longer-form generation or multi-speaker scenarios remains to be tested by the community. The open-weight release and ComfyUI support lower the barrier for experimentation.